Low-Resource Language AI: Questions to Ask Before a Project

Be specific about the language variety Confirm the language name, variety, script, region, and domain relevant to the use case. Communities may use multiple orthographic or spoken forms.

Assess what data exists Identify available sources, permissions, format, quality, and gaps. Publicly accessible material is not automatically appropriate or licensed for every use.

Include community and cultural context Where appropriate, consider consultation, consent, compensation, representation, and how the project’s purpose is communicated.

Define evaluation carefully A model or dataset should be assessed against its intended users and use cases. A single aggregate score may hide differences across dialects, domains, or speaker groups.

Document limitations Record coverage gaps, collection methods, assumptions, and known constraints. Transparent limitations help downstream users interpret results responsibly.