GAND: Gender-Ambiguous Natural Data for NLP Bias Evaluation

GAND (Gender-Ambiguous Natural Data) is a benchmarking resource for evaluating gender bias in machine translation and other NLP tasks. The dataset was compiled by jhacken from natural data resources and was last updated on 2026-05-21. It has been filtered to ensure gender ambiguity with respect to specific referents.

Use Cases

• Benchmarking gender bias in machine translation models based on gender-ambiguous natural text.
• Evaluating fairness in downstream NLP tasks based on curated ambiguous referents.
• Training or testing debiasing algorithms based on a resource compiled from OpenSubtitles and C4.

Strengths

• Data is derived from natural language sources (OpenSubtitles and C4), suggesting real-world linguistic patterns.
• Compilation involved meticulous automatic and manual filtering to ensure gender ambiguity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

OpenSubtitles from the OPUS project and C4.

Collection Method

Automatically and manually filtered from natural data resources.

Freshness

Last updated 2026-05-21 15:40:05; freshness should be verified.