Google Translate Unveils Largest Update in Translation History!
Google has announced a monumental update to Google Translate, introducing support for 110 additional languages. This expansion increases the total number of supported languages from 133 to 243, marking the most extensive update since its inception. Notably, this update includes Cantonese, a language long requested by users.
This expansion was made possible by advancements in the PaLM 2 AI language model. According to Isaac Caswell, a researcher at Google, PaLM 2 excels in learning highly correlated languages such as Awadhi and Marwadi related to Hindi, as well as creole languages like Seychellois Creole and Mauritian Creole, which are closer to the French language family.
Caswell highlighted the inclusion of Cantonese, stating it has been one of the most requested languages for Google Translate. He acknowledged the challenges posed by Cantonese’s overlap with Mandarin in written form, underscoring the difficulty in sourcing appropriate data and training models.
In addition to Cantonese, approximately a quarter of the newly added languages originate from Africa, demonstrating Google’s commitment to digitizing African languages.
Caswell revealed that most newly added languages boast at least one million users, with some having user bases in the hundreds of millions. This inclusion significantly broadens the utility of Google Translate in multilingual environments.
PaLM 2, the powerhouse behind this language expansion, enhances translation accuracy and naturalness by effectively learning and comprehending new languages and establishing connections between related languages.
PaLM 2, Google’s second-generation large-scale language model released in 2023, builds upon the strengths of its predecessor with improved capabilities in mathematical reasoning, logical inference, and coding.
Google PaLM 2 is available in four versions—Unicorn, Bison, Otter, and Gecko—catering to various device types, with the lightweight Gecko version even capable of offline use on mobile devices.
Trained on over 100 languages, PaLM 2 excels in tasks such as natural language understanding, translation, coding, question answering, summarization, creative writing, mathematical logic, and common-sense reasoning. Its semantic understanding allows it to interpret non-literal texts such as riddles and idioms.
Sources: The Verge, Google