Penn Discourse Treebank 2.0 (PDTB-2)

Intra-sentence No Polarity Signal No Strength Signal

PDTB-2.0 annotates discourse relations — including causal, temporal, and contingency senses — across the Penn Treebank WSJ corpus, using explicit connectives and implicit relation spans. It is the predecessor to PDTB 3.0.

Blocked — separate paid LDC license, no existing preprocessing recipe

PDTB-2.0 is gated behind its own paid Linguistic Data Consortium license (LDC2008T05) — distinct from PDTB 3.0's (LDC2019T05); owning one does not grant access to the other. Unlike PDTB 3.0, UniCausal has no existing preprocessing recipe for PDTB-2.0's (different, earlier) annotation format, so even with a license, someone would need to write a PDTB-2-specific converter from scratch. Not yet attempted — see resources/datasets/PDTBv2/conversion_script.py.

Overview

Domain	News / WSJ
Year	2008
Sentences	—
Granularity	Intra-sentence

Tasks & Splits

Task	Train	Test

Citation

@inproceedings{prasad:2008,
  title = {The {{Penn Discourse TreeBank}} 2.0},
  booktitle = {Proceedings of the {{International Conference}} on {{Language Resources}} and {{Evaluation}}, {{LREC}} 2008, 26 {{May}} - 1 {{June}} 2008, {{Marrakech}}, {{Morocco}}},
  author = {Prasad, Rashmi and Dinesh, Nikhil and Lee, Alan and Miltsakaki, Eleni and Robaldo, Livio and Joshi, Aravind K. and Webber, Bonnie L.},
  year = 2008,
  publisher = {European Language Resources Association},
  url = {http://www.lrec-conf.org/proceedings/lrec2008/summaries/754.html},
  urldate = {2025-05-08}
}