Google Translate can process and generate both text and voice, meaning that users can type or speak in one language, then read or listen to the translation.
“Cantonese has long been one of the most requested languages” for the service, Isaac Caswell, senior software engineer at Google Translate, wrote in a blog post. “Because Cantonese often overlaps with Mandarin in writing, it’s tricky to find data and train models.”
Cantonese and Mandarin are both written in Chinese characters, but Cantonese has some differences in grammar, terms and structure.
With the help of the US tech company’s PaLM 2 large language model (LLM), launched over a year ago, Google Translate can learn languages that are closely related to each other more efficiently, according to Caswell.
The rapid advancement of LLMs – the technology behind a new breed of intelligent AI tools such as OpenAI’s ChatGPT and Google’s Gemini chatbot – has set off a race among Big Tech companies and start-ups to roll out new service features and improve existing functions.
As Microsoft, OpenAI’s biggest investor, rushes to bake more advanced capabilities into its Edge web browser, Google on Wednesday brought five new features to its mobile Chrome browser.
They include short-cuts for local search results, a redesigned address bar, personalised search suggestions, trending search recommendations, and live sports updates.
Google Translate is not the first service to support Cantonese.
However, none of those services are as popular as Google among internet users in Hong Kong, where Google.com remained the most visited website as of last month, according to traffic tracker Similarweb.
Other new languages introduced to Google Translate include Punjabi (Shahmukhi), the most spoken language in Pakistan, as well as the Hindi-related languages of Awadhi and Marwadi.
