Meta’s AI translator has the ability to translate non-written languages

Meta’s AI translator has the ability to translate non-written languages

It is said that the AI ​​translation system of Meta, a large social networking company, can now translate languages ​​that do not have a script.

About half of the 7,000 languages ​​in the world, 4 out of 10 languages ​​are said to be undocumented.

Unstructured language poses a unique problem for modern machine learning translation systems.

Usually, machine learning translation systems before translating a new language. Before converting from text to speech, you can convert spoken words to handwritten texts.

That’s why modern machine learning translation systems are finding it difficult to translate non-written languages. The translation system developed by Meta claims to solve this problem with the latest open-source language AI upgrade.

Part of Meta’s Universal Speech Translator (UST) program, the translation system is designed to provide direct speech-to-speech translation.

As part of this project, Meta researchers worked to translate Hokkien, an unwritten language spoken throughout Asia and one of Taiwan’s official languages, into English. Machine learning translation systems are used to translate text from one language to another. You can practice the words by taking examples.

Unwritten languages ​​such as Hokkien use speech-to-unit translation (S2UT) to convert input speech into audio unit waves in a direct path, according to Mark Zuckerberg’s post.

Currently, the system helps a Hokkien speaker communicate with an English speaker. The model can only translate one full sentence at a time, and the technology could later be applied to more languages ​​and improve actual translation, he said. )

Meta has made this project open-sourced for public use, and has released the first voice-to-voice translation measurement system based on Hokkien speech.