Difference between revisions of "Paraphrase Identification (State of the art)"

From ACL Wiki
Jump to: navigation, search
Line 18: Line 18:
! Algorithm
! Algorithm
! Reference
! Reference
! Type
! Description
! Accuracy
! Accuracy
! F
! F

Revision as of 13:49, 24 March 2009

  • Microsoft Research Paraphrase Corpus (MSRP)
  • see Dolan, Quirk, and Brockett (2004)
  • train: 4,076 sentence pairs (2,753 positive: 67.5%)
  • test: 1,725 sentence pairs (1,147 positive: 66.5%)

Sample data

  • Sentence 1: Amrozi accused his brother, whom he called "the witness", of deliberately distorting his evidence.
  • Sentence 2: Referring to him as only "the witness", Amrozi accused his brother of deliberately distorting his evidence.
  • Class: 1 (true paraphrase)

Table of results

Algorithm Reference Description Accuracy F
MCS Mihalcea et al. (2006) unsupervised combination of several word similarity measures 70.3% 81.3%


Dolan, B., Quirk, C., and Brockett, C. (2004). [http://acl.ldc.upenn.edu/C/C04/C04-1051.pdf Unsupervised construction of large paraphrase corpora: Exploiting massively parallel news sources], Proceedings of the 20th international conference on Computational Linguistics (COLING 2004), Geneva, Switzerland, pp. 350-356.

Mihalcea, R., Corley, C., and Strapparava, C. (2006). Corpus-based and knowledge-based measures of text semantic similarity, Proceedings of the National Conference on Artificial Intelligence (AAAI 2006), Boston, Massachusetts, pp. 775-780.

See also