Proposition Bank I

Item Name: Proposition Bank I
Author(s): Martha Palmer, Paul Kingsbury, Olga Babko-Malaya, Scott Cotton, Benjamin Snyder
LDC Catalog No.: LDC2004T14
ISBN: 1-58563-304-6
ISLRN: 874-058-423-080-1
Release Date: September 01, 2004
Member Year(s): 2004
DCMI Type(s): Text
Data Source(s): newswire
Project(s): ACE, GALE, TIDES
Application(s): information extraction, machine translation, question-answering
Language(s): English
Language ID(s): eng
License(s): LDC User Agreement for Non-Members
Online Documentation: LDC2004T14 Documents
Licensing Instructions: Subscription & Standard Members, and Non-Members
Citation: Palmer, Martha, et al. Proposition Bank I LDC2004T14. Web Download. Philadelphia: Linguistic Data Consortium, 2004.
Related Works: View


Proposition Bank I was developed at the University of Pennsylvania's Computer and Information Sciences Department and contains semantic annotation of verbs from over one million words of English newswire text. The data was taken from the Wall Street Journal section of Treebank-2 (LDC95T7).

More specifically, each verb occurring in the Treebank was treated as a semantic predicate and the surrounding text was annotated for arguments and adjuncts of the predicate. The verbs were also tagged with coarse grained senses and with inflectional information.

All data was subject to double blind, adjudicated annotation.


There are two basic components to Propbank:

  • The Verb Lexicon. A frames file in XML format, consisting of one or more frame sets, has been created for each verb occuring in the Treebank. These files serve as a reference for the annotators and for users of the data. 3,323 such files have been created, totalling about 5.5 MB of uncompressed data.
  • The Annotation. There are approximately 113,000 annotated verb tokens. These verb tokens include all those occurring in over one million words of the Wall Street Journal section of the Penn Treebank, excluding 'be' and auxiliary uses of 'do' and 'have.' There are annotations for over 3,200 unique verbs. These annotations are stored in a single file in standoff format, totalling ~9.6 MB of uncompressed data.



Please check the Propbank homepage for updates, tools, annotation guidelines, and published papers.


This work was funded by DoD Grant MDA904-00C-2136, NSF grant IIS-9800658, and the Institute for Research in Cognitive Science at the University of Pennsylvania NSF-STC grant SBR-89-20230.

Available Media

View Fees

Login for the applicable fee