Occurred: July 2020
An asylum claim made by a Pashto-speaking Afghan refugee was denied by the US government due to an inaccurate automated translation tool.
Rest of the World reported that the individual, who had fled Afghanistan, had her asylum claim to the USA rejected because her written application did not match the story told in her initial interviews.
In the interviews, the refugee had said that she had made it through one particular event alone, but her written statement appeared to reference other people due to an automated translation tool that swapped the 'I' pronoun in the woman’s statement to 'we.'
The incident highlights the growing use of machine learning-based translation tools in immigration procedures in the USA and elsewhere, the importance of technicalities in asylum processing, and the need for system accuracy combined with human review for all languages.
According to UNICEF, Pashto is spoken by an estimated 45-55 million people in Afghanistan, Pakistan, and Iran.
Country: Afghanistan; USA
Sector: Govt - immigration
Purpose: Translate asylum claims
Technology: NLP/text analysis; Neural network; Deep learning; Machine learning; Reinforcement learning
News, commentary, analysis
Published: November 2023