---
dataset_info:
  features:
  - name: inputs
    dtype: string
  - name: choices_list
    list: string
  - name: labels
    dtype: int64
  - name: metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 14771791
    num_examples: 14190
  - name: validation
    num_bytes: 2083302
    num_examples: 2020
  - name: test
    num_bytes: 3857360
    num_examples: 3610
  download_size: 4378934
  dataset_size: 20712453
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: other
source_datasets: []
---

# mcscript2

MCScript2.0 native two-answer reading comprehension, original XML partitions.

Preserve stories, questions and native choices; exactly one gold answer and
two distinct non-empty options are required. Story hashes prevent train/dev/
test copies. Source URL/archive hash and per-split counts are recorded.
Dataset terms: unspecified until original distribution terms are verified.

Original data: [original distribution](https://fedora.clarin-d.uni-saarland.de/sfb1102/MCScript-2.0.zip). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/dialogue.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/dialogue.py#L50).

[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.