Facebook, Instagram and WhatsApp’s parent company, Meta, has announced a new project that it thinks will serve the company’s metaverse plans. The company is working on an artificial intelligence-based universal speech translation system that will allow all people in the world to communicate with each other, thus facilitating interaction in the digital universe.
Data scarcity is a major problem for translation systems
Meta announced two different long-term approaches to AI-assisted translation that they will showcase. The company will first create artificial intelligence models that can learn to translate the language using fewer training examples, with the system it calls No Language Left Behind. Mainly, this system will be developed for less common languages. With the second system, Universal Speech Translator, Meta will focus on translating speech from one language to another in real time.
The company also talked about three different challenges they will face while developing these two systems. The first of the difficulties mentioned is the insufficient dataset for machine translation (MT) to translate all languages between each other. The other challenge will be modeling challenges that arise as systems grow with more languages. Meta also says they need to find new ways to evaluate and improve results.
Meta blog post: “MT systems for text translations typically rely on learning from millions of sentences of annotated data. Therefore, MT systems capable of high-quality translation have only been developed for a few web-dominated languages. To other languages. Expanding means finding ways to acquire and use educational examples from languages with sparse web presence.” statements were included.
Aiming for a more inclusive and connected world by realizing both of the mentioned systems, Meta brings together the digital and physical worlds in the metadata base with people they don’t know like chatting with their neighbors. states that they can interact easily.
- Home
- Internet
- Social Media News
- Meta develops translation system for all spoken languages
14784
