Advancing Data Equity: Practitioner Responsibility and Accountability in NLP Data Practices

Practitioners face real conflicts between profit and fairness: The study found "persistent tensions between commercial objectives and equity commitments," suggesting that NLP workers want to build fairer systems but encounter organizational pressure to prioritize speed and cost over equity considerations.

Current diversity efforts may be superficial: The research critically examines "diversity washing" in NLP data practices, indicating that simply adding diverse data without structural changes is insufficient for achieving genuine equity in AI systems.

Practitioners want more say in decision-making: Findings call for "more participatory and accountable data workflows," meaning those directly building datasets should have greater input into how data is collected, labeled, and used—rather than decisions being made solely by management or distant policymakers.

Governance frameworks need to support workers, not just audit algorithms: The study links practitioner experiences to "multi-scalar AI governance," arguing that improving NLP equity requires structural reforms that give practitioners more agency and support, alongside policy changes like the U.S. AI Bill of Rights.

Community consent is missing from current practices: The research emphasizes that "improving NLP equity requires...community consent," highlighting that the people whose data is used and who are affected by NLP systems should have a voice in how their information is handled.