My Voice, My Words: NLP and AI for more effective analysis of qualitative data from communities
Challenge
In the Kurdistan Region, UNHCR receives a lot of information from communities via unstructured text and voice data in Kurdish and Arabic (depending on the collection resource). These have to be manually processed and assessed, which is time consuming – so the data is often not fully analyzed.
Solution
Test ways that natural language processing (NLP) tools – such as translation, summarization, question response, categorization, transcription, and more – could be used to support the processing and analysis of unstructured data in languages used by communities.
Impact
Quicker and more comprehensive processing and analysis of UNHCR participatory data, freeing up UNHCR resources and enabling more effective responses. The proposed Kurdish NLP methods could also be integrated into other enterprise tools, expanding that tool’s capabilities.
Project impact
Other impact
UNHCR collects large volumes of qualitative protection data through interviews and focus groups in low-resource and minority languages, which are the languages most spoken by forcibly displaced communities. This makes manual transcription slow, error-prone and leaves refugee voices underused in decision making. Funded by the Data Innovation Fund and delivered jointly by UNHCR Iraq, RBAP DIMA, Global Data Service, and Innovation, in partnership with KoBo Inc. and CLEAR Global as linguistic and low-resource language experts, the My Word My Voice project develops the first small language models SLMs) for Sorani Kurdish and Rohingya, two languages largely unsupported by mainstream foundational and commercial models, embedded natively into KoboToolbox so audio responses are automatically transcribed and analysable at scale. The approach combines cross-lingual transfer learning, synthetic data augmentation and community-validated transcription, offering a first-of-its-kind blueprint for inclusive, language-equitable AI in humanitarian operations. The project is featured as a global use case in the ITU AI for Good "Innovate for Impact" 2025 report, positioning UNHCR as a leader in responsible AI for underrepresented languages. Lear more about Roshna, a member of the project team, and the project itself in UNHCR's Humanitarians at Work LinkedIn newsletter.