BECauSE 2.0

Intra-sentence No Polarity Signal No Strength Signal

BECauSE 2.0 annotates causal relations (three types: Consequence, Motivation, Purpose) alongside seven non-causal "overlapping" relations that share causal-like constructions (e.g. Correlation, Hypothetical) in English news text, making it one of the most relation-dense corpora for causal NLP.

The original annotation also tags each causal instance's Degree as Facilitate or Inhibit, but Inhibit ("the cause suppresses the effect") is negative-valence causation, not a denial that a causal relationship holds at all — so it isn't represented as Relation.Countercausal here (see the dataset's README for the full reasoning).

Overview

Domain	News
Year	2017
Sentences	1,803
Granularity	Intra-sentence

Tasks & Splits

Task	Train	Test
Causality Detection	453	17
Causal Event Candidate Extraction	453	17
Causality Identification	453	17
Causality Extraction	—	—

Data Explorer

Causality Detection

Causal Event Candidate Extraction

Causality Identification

Citation

@inproceedings{dunietz:2017,
  title      = {The {{BECauSE Corpus}} 2.0: {{Annotating Causality}} and {{Overlapping Relations}}},
  shorttitle = {The {{BECauSE Corpus}} 2.0},
  booktitle  = {Proceedings of the 11th {{Linguistic Annotation Workshop}}, {{LAW}}@{{EACL}} 2017, {{Valencia}}, {{Spain}}, {{April}} 3, 2017},
  author     = {Dunietz, Jesse and Levin, Lori S. and Carbonell, Jaime G.},
  editor     = {Schneider, Nathan and Xue, Nianwen},
  year       = 2017,
  pages      = {95--104},
  publisher  = {Association for Computational Linguistics},
  doi        = {10.18653/V1/W17-0812},
  urldate    = {2025-05-08}
}