Google Translation’s Largest Update in History!
Google announced that Google Translate has added support for 110 new languages, increasing the total number of supported languages from 133 to 243. This includes Cantonese, which users have been requesting for years, making it the largest update since the launch of Google Translate!
According to Google, this update is made possible by the PaLM 2 AI language model. Isaac Caswell, a Google research scholar, pointed out that the PaLM 2 AI model performs exceptionally well in learning highly correlated languages. For example, it excels in languages like Awadhi and Marwadi, which are related to Hindi, as well as French-based creole languages like Seychellois Creole and Mauritian Creole.
Google now understands Cantonese too. Among the newly supported languages, Google Translate now supports Cantonese. Caswell stated that Cantonese has always been one of the most requested languages for Google Translate. However, due to the overlap between Cantonese and Mandarin in writing, finding suitable data and training models has been a challenge.
Caswell also mentioned that approximately a quarter of the new languages added come from Africa, demonstrating Google’s commitment to promoting digitalization of African languages. He revealed that most of these new languages have at least one million users, and some languages have hundreds of millions of users. This expansion of languages broadens the usage of Google Translate and enhances its practicality in multilingual environments.
So, what is PaLM 2 and how powerful is it? Google stated that the technology behind the language expansion in this update primarily relies on the strong learning ability of the PaLM 2 AI language model. This model not only effectively learns and understands new languages but also establishes connections between related languages, thereby improving translation accuracy and naturalness.
PaLM 2 is the second-generation Large Language Model (LLM) released by Google in 2023. Compared to the first-generation PaLM, the second-generation significantly enhances its capabilities in areas such as mathematics, logical reasoning, and encoding.
Google PaLM 2 can be divided into four versions: Unicorn, Bison, Otter, and Gecko, from largest to smallest. They can be used on different types of devices, and the lightweight Gecko version can even be used offline on mobile devices.
Google stated that PaLM 2 has been trained using over 100 languages and excels in understanding and generating natural language, translation, encoding, question-answering, summarization, creative writing, mathematical logic, and common-sense reasoning. In particular, PaLM 2 can comprehend non-literal meanings in texts, such as riddles and idioms.
Source: The Verge, Google
Further reading:
How has NVIDIA been consistently successful? What is their quantum computing layout? Why did Jensen Huang emphasize two key words at the shareholders’ meeting?
This article was initially written by an AI and edited by Li Xiantai.