Meta has released its first multimodal, multilingual AI translation and transcription model named SeamlessM4T. It claims the model can translate over 100 languages.
SeamlessM4T can recognise the speech of over 100 languages, as well as translate from spoken word to text. Likening the system to the fictional Babel Fish in The Hitchhiker’s Guide to the Galaxy, Meta has described the software as a universal language translator.
The company has also released SeamlessM4T for free under a research license in an effort to uphold its commitment to open science.
The metadata SeamlessM4T was trained on will also be publicly available to researchers and is the biggest multilingual dataset to date, totalling approximately 270,000 hours of mined text and spoken word.
Meta’s release of such a dataset is significant since previous research has suggested that large language models and other AI chatbots may respond to English prompts better because they were mainly trained on English-language data.
GlobalData senior analyst Beatriz Valle described the release as great news.
How well do you really know your competitors?
Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.
Your download email will arrive shortly
Not ready to buy yet? Download a free sample
We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below formBy GlobalData
“With AI taking over the market, the dominance of English has become apparent,” Valle explains, “and creating robust models in a variety of languages is a social and commercial imperative.”
Whilst more accurate translation tools are especially useful for social media companies like Meta in moderating content in different languages, Valle also believes this release will spur on Meta’s competitors to fulfil a “glaring gap in the market”.
Meta’s open-source release of the software and its metadata to researchers is an “interesting” tactic in Valle’s opinion.
She speculates that this choice could be an attempt by Meta to pre-emptively “diminish” other multilingual platforms in the market. However, the move will also ignite innovation in the sector as well.
Nevertheless, Valle concludes, this marketing strategy works for Meta.
The release will have to balance between Meta’s ability to reap the benefits of the free work the release will allow software developers to carry out, whilst also posing significant risks in both AI safety and copyright infringement.
Meta stated that it hopes the release of SeamlessM4T will help lead the world to a future where everyone can be better understood.
However, in light of its recent backtrack to provide Llama2 for free, some may be sceptical of the longevity of SeamlessM4T’s availability.