After Amharic, Hausa and Swahili, ten new African languages can now be translated by the Google Translation service. Alongside Asian or South American languages, such as Bhojpuri (India, Nepal, Fiji), Dhiveri (Maldives) or Quechua (Peru), a handful of new idioms, among the approximately 2,000 used on the African continent, are therefore made available to Internet users.
This free service offered by Google allows you to instantly translate sentences and entire texts into 133 different languages. Bambara (Mali), Ewe (Ghana, Togo), Krio (Sierra Leone), Lingala (Central Africa), Luganda (Uganda, Rwanda), Oromo (Ethiopia), Sepedi ( South Africa), Tigrinya (Eritrea, Ethiopia), Tsonga (South Africa) and Twi (Ghana). Over 300 million people are fluent in these 24 new idioms available for translation.
Google prides itself on helping to “break down language barriers” and “connect communities around the world”, ensuring support for peoples whose languages remain underrepresented in most technologies. “Google Translate users have never been so numerous, but we still have work to do to make it universal and accessible to everyone,” the tech giant’s CEO, Indian Sundar Pichai, said on May 11. . He was speaking at “Google I/O,” an annual two-day conference hosted by Google at the Moscone Center in San Francisco, California.
This innovation represents a “significant new step” in the technology used by Google Translate, insisted Isaac Caswell, a Google engineer. “These languages are the first that have been added via the ‘Zero-shot’ translation software,” explained the engineer. This learning model allows this machine to translate a language without drawing inspiration from an existing example.
The languages represented are mainly European and neglect linguistically very varied regions, such as Africa
Originally, to develop the Google Translate service, the company used bilingual dictionaries. A costly and time-consuming approach, which will later be replaced by statistical machine translation, based on a large database. For these 24 new languages, the artificial intelligence machine therefore makes it possible to translate texts, even without a prior database.
This innovation contributes to “the extension of coverage to many communities that were very neglected, not only by Google but also by other technologies in general”, added Isaac Caswell.
Neglected African languages
Google has also relied on people whose mother tongues these 24 new idioms are. “Their help has been fundamental in our work”, insisted the company, which has collaborated with more than 100 people and institutions.
“Although translation services cover the most widely spoken languages in the world, they only include about 100 in total, or only 1% of all existing languages,” Google conceded. Moreover, the languages represented are mainly European and neglect linguistically very varied regions, such as Africa or the Americas. »