After graduating in English in 2014, Bokan Jaff struggled to find steady work and took translation jobs in Halabja. Neighbors kept asking why Persian, Arabic, English, and other languages were available on automated platforms while Sorani was not. Sorani's Arabic-like script also made it harder for systems trained on Latin alphabets.

In 2019, an Argentine friend connected Jaff to Google's Crowdsource work. Jaff became the bridge between Google engineers and Kurdish language enthusiasts: he coordinated translations, checked other people's work, and helped fix Sorani-specific problems. When a beta was ready in May 2021, he sent it to the Google Crowdsource Facebook group; about 5,000 users responded, and volunteers tested it for a year.

Google launched Sorani alongside 23 other languages in May 2022. The service became widely available to an estimated 8 million Sorani speakers, and former Google engineer Keith Stevens attributed its quality to the community's human validation. The result was meaningful but imperfect: Google owned the service, volunteers were unpaid, and a politically sensitive translation bug remained.

Google launched Sorani support for an estimated 8 million speakers; a volunteer network supplied the human validation needed to make the beta widely available.

Sorani is used by an estimated 8 million people but was missing from major automated translation platforms.

Its Arabic-derived script posed a practical processing challenge for systems trained on Latin-alphabet languages.

Google needed enough native-speaker evaluation to decide whether the model was accurate and safe to launch.

Jaff already had a community he could mobilize, giving Google a low-cost source of language-specific validation.

When a language or market lacks conventional data, the missing asset may be a committed user community. Build and coordinate that community before asking the product to scale.

The launch made Sorani available globally and turned Jaff into a local symbol celebrated by media, politicians, diplomats, and ordinary people. The service still had bugs, including occasional confusion between rival Kurdish political parties; Jaff and a former Google engineer also criticized Google for not paying contributors or releasing the data openly.

FOLLOW THE EVIDENCE

The sources

  1. These volunteers taught Google Translate to read Sorani Kurdish restofworld.org