Google Translate is adding over 100 new languages to its language repository, bringing the total number of supported languages to nearly 250.
Google launched Google Translate in 2006. Since then, the company has gradually added support for more languages to this platform. In May, the company added support for 24 new languages, increasing the total number of supported languages to 133.
However, the update on June 27 could be considered the largest update ever for Google Translate. The tech giant announced that the tool is undergoing “the biggest expansion ever”, with 110 new languages, bringing the total number of supported languages to 243.
For comparison, Apple Translate supports 20 languages and Microsoft Translator supports 135 languages. Currently, Google Translate is leading the way, far ahead of these two competitors.
According to Google, this achievement was made possible thanks to the large language model PaLM 2. Some of the newly supported languages include Afar, Cantonese, Manx, NKo, Punjabi, Tamazight, and Tok Pisin.
Google achieved this milestone thanks to the PaLM 2 model. (Photo: Google).
The company stated that the 110 new languages added “represent over 614 million speakers, opening up translation possibilities for about 8% of the world’s population.”
Additionally, “Some of these are major languages spoken by over 100 million people. Others are used by small indigenous communities, and a few have very few native speakers but are in the process of being revived.”
A quarter of the 110 new languages come from Africa, making this the largest update of Google Translate for the residents of the continent. With 243 languages, the total number of translation pairs in Google Translate has now reached 29,403.
What makes the PaLM 2 model special?
PaLM stands for Pathways Language Model, which utilizes Google’s Pathways AI architecture to train machine learning models to perform various tasks. This is a crucial part of Google’s vision to achieve AGI – artificial general intelligence – the goal for AI to perform all cognitive tasks that humans can do.
Google has previously stated that the Pathways architecture will pave the way for multimodal AI beyond text, and Gemini is another large language model (LLM) that has achieved this goal. In a blog post, the company mentioned:
Pathways can enable multimodal models to understand vision, audio, and language simultaneously. Therefore, whether the model is processing the word “leopard,” the sound of someone saying “leopard,” or a video of a leopard running, a consistent response will be activated from within: the concept of a leopard. As a result, it is a deeper model that makes fewer mistakes and has less bias.
Unlike the LaMDA model it replaces, PaLM 2 has been trained on over 100 languages and even has better domain knowledge in fields such as coding. It also has significantly improved logical reasoning and mathematical capabilities.
To support Google Translate, PaLM 2 has been trained on a large volume of parallel multilingual text. This model can learn and translate languages faster thanks to its logical capabilities, especially for languages that are closely related, such as those in the Hindi group.
Google aims to support the 1,000 most widely used languages in the world, in order to reduce language barriers and promote global community connectivity.
