metadata
dataset_info:
features:
- name: sentence1
dtype: string
- name: sentence2
dtype: string
- name: labels
dtype: string
- name: metadata
dtype: string
splits:
- name: train
num_bytes: 5405348
num_examples: 14000
- name: validation
num_bytes: 767748
num_examples: 2000
- name: test
num_bytes: 1535708
num_examples: 4000
download_size: 3561518
dataset_size: 7708804
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
license: cc0-1.0
source_datasets: []
edos
EDOS native binary sexism judgments and original train/dev/test assignments.
Original source: https://github.com/rewire-online/edos (CC0-1.0). Reads the aggregated annotation CSV, retaining all native labels, identities and split assignments. Category/vector labels remain in metadata; no generated explanations enter inputs. See provenance.json for the pinned source/hash.
Original data: original distribution. Repackaged as parquet for tasksource by scripts/repackage_dataset/specialized_text.py.
Conversion provenance: source revisions, hashes, configuration and exclusions.