In May 2022, Google added 24 new languages spoken by over 300 million people worldwide to Google Translate—marking the first time the platform used Zero-Shot Machine Translation trained purely on monolingual text without bilingual parallel corpora.

Overcoming the Scarcity of Bilingual Parallel Text

At Google I/O in May 2022, Google announced the addition of 24 under-resourced languages to Google Translate, bringing its total supported languages to 133. Among the newly added languages are Indigenous languages of the Americas (Quechua, Guarani, Aymara), African languages (Twi, Bambara, Ewe, Lingala, Oromo, Tigrinya), and South/Southeast Asian languages (Assamese, Bhojpuri, Sanskrit, Mizo). Historically, adding a language to an NMT system required millions of bilingual translated sentence pairs—data that simply does not exist online for many regional languages.

How Zero-Shot Monolingual Learning Works alongside Native Speaker Validation

To bridge this gap, Google researchers utilized Zero-Shot Machine Translation, where a massive multilingual neural network learns to translate into a target language after seeing only monolingual text in that language, transferring structural translation knowledge acquired from over 100 high-resource languages. Crucially, Google collaborated directly with native speakers, linguists, and academic institutions to curate evaluation benchmarks and filter noisy web crawls—demonstrating how AI and human linguistic expertise complement one another.

Avete bisogno di traduzioni multilingue certificate, asseverazioni legali o localizzazione conforme agli standard ISO per i vostri progetti internazionali? Richiedete un preventivo gratuito personalizzato ad Alphatrad e riceverete un piano operativo dettagliato entro 24 ore.