According to reports, the large social network company Meta’s artificial intelligence translation system is now capable of translating languages with no written script.
Four out of every ten languages, or about half of the world’s 7,000 languages, are said to be undocumented. Modern machine learning translation systems face a unique challenge when dealing with unstructured language. Usually, before translating into a new language, machine learning translation systems. You can convert spoken words to handwritten texts before converting from speech to text.
Non-written languages are challenging for modern machine learning translation systems because of this. With the most recent open-source language AI upgrade, Meta claims to have developed a translation system that addresses this issue. The translation system is intended to provide direct speech-to-speech translation and is a component of Meta’s Universal Speech Translator (UST) program.
Meta researchers worked on the English translation of Hokkien, one of Taiwan’s official languages and an unwritten language spoken throughout Asia. Text is translated from one language to another using machine learning translation systems. By using examples, you can practice the words. Unwritten dialects, for example, Hokkien use discourse to-unit interpretation (S2UT) to change over input discourse into sound unit waves in an immediate way, as per Imprint Zuckerberg’s post.
The system currently facilitates communication between English speakers and Hokkien speakers. The model can decipher each full sentence in turn, and the innovation could later be applied to additional dialects and work on real interpretation, he said. Meta has released the first voice-to-voice translation measurement system based on Hokkien speech and made this project open-sourced for public use.