The reason for this study is the technical details of the universal speech translator and the existence of an “artificial intelligence plan” that Meta Al has tried to develop for an embodied interactive metaverse. Announcing that it is working on the artificial intelligence-based "universal speech translator" (IHR) in February 2022, Meta Al announced that it was working on Urdu, Luganda and Asturian languages, etc., with the content of the "No Language Left Behind" (NLLB) project in July 2022. Meta Al also accelerated its work for the translation of unwritten spoken languages. In this case, the answer for the question “How can non-written spoken language(s) be translated?” arises.
Why did meta focus on unwritten languages?
In line with the information Meta announced, the languages spoken by 20 (twenty) percent of the world's population are not compatible with the translation tools of languages such as English, Mandarin and Spanish. Meta states that the artificial intelligence-assisted translation tools in use today focus on spoken written languages and are in this direction. In line with this information, more than 40 percent of the spoken languages in the world are not included in the translation content.
Technical details of universal speech translator
For this purpose, Meta Al strives to overcome the difficulties of 3 (three) different translation systems. For this, the company obtained more input data in line with the model training in more languages. Thus, it can be said that the company has developed many different methods of utilizing the existing and obtained data. A problem arises here: as models evolve, new modeling methods need to be developed so that more languages can be translated. The company actively carries out new studies for this purpose. The company's research and development team need a sample "case" study to be able to compare data sets from training data to their modeling methods.
Referance:
https://webrazzi.com/2022/10/21/meta-ai-yazili-olmayan-konusma-dilini-nasil-tercume-ediyor
