During the 1960 Paramount News feature, “Electronic ‘Brain’ Translates Russian to English”, the presenter ends the discussion of the machine’s capabilities with a final question: “Is this the end of human translators?” The researcher responds yes, at least in the realm of translating scientific and technical material.
It’s been 60 years since then, but human translators are still needed in different fields. This goes to show that artificial intelligence has yet to reach a level of accuracy in translation deemed enough to replace humans. The problem was that AI didn’t understand context, adaptation, and differences between languages. However, with the boom of AI, the gap between human and AI skills is narrowing. Enter Meta’s SeamlessM4T.
Bridging gaps between languages
SeamlessM4T is a versatile, multimodal, and multilingual AI translation model that supports speech-to-speech, speech-to-text, text-to-speech, and text-to-text translation. Whether you’re speaking, typing, or reading, SeamlessM4T aims to bridge the gap between languages.
There have been attempts to create this, but the AI back then had problems with producing comprehensive speech in other languages. Today, SeamlessM4T’s support for speech-to-speech translation can be a game-changer. That’s because it allows individuals speaking different languages to engage in real-time conversations without the need for an intermediary.
Meta’s strategy
SeamlessM4T is building upon Meta’s prior advancements in AI language technology. This includes the No Language Left Behind (NLLB) model and the Universal Speech Translator. The model is also powered by SeamlessAlign, the largest open speech-to-speech and speech-to-text with over 4 million hours of recorded speech.
Instant real-time translation over VOICE, TEXT, and IMAGES
Communication Barrier Soon to be Eradicated Indefinitely🗣️
a SeamlessM4T multimodule translation model will be available
But how will Meta break the barrier?
This is how you will be able to leverage this new tech🧵👇 pic.twitter.com/kwpGgBldnB
— Eder Teixeira 🍭 (@EderTeixeira333) August 24, 2023
As a testament to Meta’s open science commitment, SeamlessM4T was released to the research and developer community under the research license. Researchers and developers worldwide can build upon it, creating applications and solutions that transcend language barriers together. In fact, you can try the demo of the model right now.
The implications of SeamlessM4T extend to numerous domains, from education and business to diplomacy and entertainment. While it may take a while for people to realize or to start trusting AI, human translators seem to be less needed. It took over 60 years, but those researcher’s words seem to be finally coming to fruition.
Photo credit: The feature image is a collage made of the material from the press video and is owned by Meta.
Source: NEC / Meta
