Google Translate Adds 110 New Languages

From Cantonese to Qʼeqchiʼ, these new languages cover over 614 million speakers, opening up translations for about 8% of the world's population. Some are big world languages with over 100 million speakers, while others are spoken by small Indigenous communities. A few are even being actively revived despite having almost no native speakers left. About a quarter of these new languages are from Africa, marking our largest expansion of African languages yet, including Fon, Kikongo, Luo, Ga, Swati, Venda, and Wolof.

Here are some of the newly supported languages in Google Translate:

• Afar, a tonal language spoken in Djibouti, Eritrea, and Ethiopia, had the most volunteer community contributions of all the languages in this launch.
• Cantonese, long one of the most requested languages for Google Translate, is tricky to handle because it often overlaps with Mandarin in writing, making it hard to find data and train models.
• Manx, the Celtic language of the Isle of Man, was nearly extinct after the last native speaker passed away in 1974. But thanks to a revival movement, there are now thousands of speakers.
• NKo, a standardized form of the West African Manding languages, unifies many dialects into a common language. Its unique alphabet was invented in 1949, and it has an active research community today.
• Punjabi (Shahmukhi), the variety of Punjabi written in Perso-Arabic script, is the most spoken language in Pakistan.
• Tamazight (Amazigh), a Berber language spoken across North Africa, has many dialects but a generally mutually understandable written form. It's written in both Latin and Tifinagh scripts, both supported by Google Translate.
• Tok Pisin, an English-based creole and the lingua franca of Papua New Guinea, might be understandable to English speakers when translated.

How we choose language varieties

Adding new languages to Translate involves a lot of thought, from choosing which varieties to include to deciding on specific spellings.

Languages have tons of variation: regional varieties, dialects, different spelling standards. Many don't have a single standard form, so picking the "right" variety is tough. We focus on the most commonly used varieties. For example, Romani has many dialects across Europe. Our models produce text closest to Southern Vlax Romani, a commonly used variety online, but also mix in elements from others like Northern Vlax and Balkan Romani.

PaLM 2 was crucial in helping Translate learn languages that are closely related, like those near Hindi, such as Awadhi and Marwadi, and French creoles like Seychellois Creole and Mauritian Creole. As tech advances and we keep working with expert linguists and native speakers, we'll support even more language varieties and spelling conventions over time.

Check out the Help Center to learn more about these newly supported languages. And start translating at translate.google.com or on the Google Translate app on Android and iOS.