TCR

Inter-sentence (whole document/section) No Polarity Signal No Strength Signal

TCR (Temporal and Causal Reasoning) provides joint temporal and causal relation annotations over a small but densely annotated set of news articles, enabling reasoning that combines both relation types.

Extraction and identification rows are whole documents (see the granularity pill above): these 25 news articles were annotated specifically for long-range, cross-sentence causal reasoning, so a per-sentence schema would drop or corrupt a large share of the real links. Detection is derived at sentence granularity instead (via causalatee.data.utils.identification_batch_to_detection_sentences), since every document here was selected for containing causal chains and a whole-document label would otherwise never be negative.

TCR shares its 25 source documents with EventCausality, but the two are not the same annotations — TCR re-extracted events (ClearTK) and independently re-annotated/pruned causal links (580 → 172) for a different paper. Don't conflate results across the two.

Overview

Domain	News
Year	2018
Sentences	172
Granularity	Inter-sentence (whole document/section)

Tasks & Splits

Task	Train	Test
Causality Detection	—	—
Causal Event Candidate Extraction	—	—
Causality Identification	—	—
Causality Extraction	—	—

Data Explorer

Causality Detection

Causal Event Candidate Extraction

Causality Identification

Citation

@inproceedings{ning:2018,
  title     = {Joint {{Reasoning}} for {{Temporal}} and {{Causal Relations}}},
  booktitle = {Proceedings of the 56th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}}, {{ACL}} 2018, {{Melbourne}}, {{Australia}}, {{July}} 15-20, 2018, {{Volume}} 1: {{Long Papers}}},
  author    = {Ning, Qiang and Feng, Zhili and Wu, Hao and Roth, Dan},
  editor    = {Gurevych, Iryna and Miyao, Yusuke},
  year      = 2018,
  pages     = {2278--2288},
  publisher = {Association for Computational Linguistics},
  doi       = {10.18653/V1/P18-1212},
  urldate   = {2026-06-10}
}