SeamlessM4T AI system by Meta will help users translate 100 languages

If we say that Artificial Intelligence is the most popular domain in 2023, then it won’t be wrong. Tech companies are working hard to introduce the most advanced AI systems. In line with these efforts, Meta has come forward with a new AI model termed SeamlessM4T. By accurately translating and transcribing approximately 100 languages in both written and spoken forms, it hopes to dissolve linguistic barriers.

The company states that the standard language translation tools have several limitations. Comparatively, the SeamlessM4T has the ability to deal with speech-to-text and text-to-text translations. It can process input from 100 languages and produce 35 different output languages. Furthermore, communicating in several languages is made simple by the language model’s ability to complete all of the aforementioned tasks in a single step.

Meta is utilizing a mechanism to filter out negative or abusive words in order to ensure sensitivity in translations. “In training data, we filtered out imbalanced toxicity. We deleted that training sequence when input or output had varying levels of toxicity,” said Meta.

Avoiding gender bias

The features and capabilities of the new SeamlessM4T AI system are praiseworthy, but it is still prone to inherent gender biases. Several other translation tools, like Google Translate, have already been affected by this. To resolve these issues, the company has introduced a system that will identify and eliminate gender bias in translations.

Despite its remarkable talents, the AI model still finds it difficult to accurately reproduce the nuanced lexical variations found in human translations—a crucial component in meaningful interactions. Additionally, Meta has stressed that the model is inappropriate for use in medical or legal contexts. Furthermore, the company doesn’t recommend this tool for lengthy translations.

“This single system design decreases errors and delays, improving the speed and caliber of the translation process, and moving us closer to the realization of seamless translation. Juan Pino, a research scientist at Meta’s AI research division, said, “In the near future, we aim to study how this fundamental model may facilitate new communication skills, ultimately moving us closer to a society where everybody can be understood”.