Global PIQA
Contributing Zarma examples to a multilingual benchmark of commonsense reasoning across languages and cultures.
My contribution
Contributing author · Zarma dataset contribution
The context
Commonsense benchmarks need examples that make sense across languages and cultural settings. My contribution focused on Zarma, a low-resource West African language, within the wider Global PIQA collaboration.
How I worked
- Prepare the examples. Worked with M. K. Keita on prompts for generating candidate examples.
- Review and curate. Helped filter and format generated examples, and reviewed French seed examples for cultural relevance.
- Contribute within a shared benchmark. Focused on the Zarma dataset rather than claiming responsibility for the full multilingual benchmark or its evaluation pipeline.
Deliverables & takeaways
I contributed to the Zarma dataset and am listed as a contributing author on the public paper. The paper provides the research questions, overall benchmark design and collaboration-wide results.
This work connects my interests in language and AI with practical dataset preparation: an example needs to be understandable and culturally relevant, not simply well formatted.
Scope & limitations
My individual contribution concerns Zarma data preparation and review. Overall benchmark results belong to the full research team. The linked preprint and its author list provide public evidence; the NeurIPS 2026 acceptance is reported in my CV.