OpenSubtitles Inferred: Movie Dialogue Annotated for Multi-Dimensional Gender Bias

A text dataset from the Multi-Dimensional Gender Bias Classification framework, based on English subtitles from OpenSubtitles. The data is automatically annotated for gender along the ABOUT, TO, and AS dimensions using inferred character gender and a classifier. The dataset was contributed by authors Xiang Zhang, Junbo Zhao, and Yann LeCun.

Use Cases

• Train a text classifier for gender bias detection based on the annotated ABOUT, TO, and AS dimensions.
• Benchmark model performance against the reported 67.13% accuracy achieved by a Transformer model on binary gender prediction tasks.
• Analyze gender representation and conversational dynamics in movie subtitles.
• Study the application of automated annotation methods, such as gender inference from kinship terms and name probability distributions.

Strengths

• Part of a documented framework for multi-dimensional gender bias classification.
• Annotations are derived from a method using gender kinship terms and U.S. baby name statistics.
• Includes binary and ternary labels with associated confidence scores.

Limitations

• Row count, column definitions, and file formats are unknown, limiting suitability assessment.
• Annotations are automatically inferred, which may introduce errors compared to human labeling.
• Last update date is unknown; freshness is unverified.

Provenance

Source

OpenSubtitles, processed by the authors of the Multi-Dimensional Gender Bias Classification project.

Collection Method

English subtitles were filtered and annotated using inferred character gender and a classifier trained on other datasets.

Time Range

null

Freshness

null

Geography

null