Usually, the developer of the translation system trains the AI model on written text in a large number of target languages. This provides the system with many different language examples and combinations to produce the most correct results.
Generally speaking, AI-driven translation systems have been greatly improved in recent years. It is now easier than ever to get translation help online or on your device for better international communication.
But one problem with such systems is that there is a delay in the translation process. For example, when the system records spoken pronunciation, the word is first converted to into text, and then translated by the AI system. The translated words are then converted back to the pronunciation so that they can be heard. The unusual thing about the
Meta project is that the development team does not have a large amount of Minnan text text to input into the artificial intelligence system.
Fujian dialect is a version or dialect of Chinese. Millions of people speak this language in Fujian Province in southeast China. Many people in Taiwan, as well as some communities in Malaysia , Singapore and Philippines , also use this language. This language is often passed down from generation to generation.
Meta points out that Fujian dialect is one of nearly 3,500 existing languages that are mainly used and do not have widely used writing systems. The company said its AI developers’ goal is to create voice-to-speech translation tools that cover most languages in the world.
Earlier this year, Meta announced two new AI projects. One is called "not forgetting language". Zuckerberg said in a video that efforts aim to create a translation system covering “hundreds of” world languages.
Another type is called a universal voice translator. The company said in a statement that the project’s goal is to create a system that can produce “vocabulary-to-vocabulary translations in all languages.” Meta's latest system involving Fujian dialect was developed as part of its universal voice translation project.
Zuckerberg said: "Being able to communicate with anyone in any language - this is the superpower people dream of, and AI will achieve this in our lifetime."
Meta said it uses several different methods to create new systems to translate Minnan and English. The team trained its AI model on a written text example from another version of Chinese Mandarin (similar to Fujian dialect).
In addition, Meta developers used a encoding tool designed to compare spoken Fujian dialect with similar English texts. The team also works closely with Minnan speakers to ensure the results are correct.
Meta says its goal is to create a voice-to-voice translation system for more languages in the future using the same method as Fujian dialect.
However, the company said its Minnan translation model is "still in progress". It points out that the system currently only translates one complete sentence at a time.
