Penn Discourse Treebank 3.0 (PDTB 3.0)

Intra-sentence Polarity Documented, Not Converted Strength Documented, Not Converted

PDTB 3.0 annotates discourse relations — including causal, temporal, and contingency senses — across the full Penn Treebank WSJ corpus using both explicit connectives and implicit relation spans.

The real PDTB-3 sense hierarchy distinguishes Contingency.Negative-condition (polarity) and Cause+Belief/Cause+SpeechAct (epistemic strength), but this project's converter only reads UniCausal's binary pair_label, so both signals are documented here without being surfaced in the converted data.

Requires your own paid LDC license — not redistributed by causalatee

PDTB 3.0 is a paid Linguistic Data Consortium resource (LDC2019T05) — distinct from PDTB 2.0's (LDC2008T05, see PDTB-2); owning one does not grant access to the other. causalatee does not fetch or host the underlying text — obtain your own license, follow UniCausal's preprocessing recipe to produce pdtb_train.csv/pdtb_test.csv, then run resources/datasets/PDTB3/conversion_script.py yourself.

Overview

Domain	News / WSJ
Year	2019
Sentences	—
Granularity	Intra-sentence

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

@article{webber:2019,
  title   = {The Penn Discourse Treebank 3.0 Annotation Manual},
  author  = {Webber, Bonnie and Prasad, Rashmi and Lee, Alan and Joshi, Aravind},
  year    = 2019,
  journal = {Philadelphia, University of Pennsylvania},
  volume  = {35},
  pages   = {108}
}