Creating disjunctive logical forms from aligned sentences for grammar-based paraphrase generation

  • Authors:
  • Scott Martin;Michael White

  • Affiliations:
  • The Ohio State University, Columbus, Ohio;The Ohio State University, Columbus, Ohio

  • Venue:
  • MTTG '11 Proceedings of the Workshop on Monolingual Text-To-Text Generation
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a method of creating disjunctive logical forms (DLFs) from aligned sentences for grammar-based paraphrase generation using the OpenCCG broad coverage surface realizer. The method takes as input word-level alignments of two sentences that are para-phrases and projects these alignments onto the logical forms that result from automatically parsing these sentences. The projected alignments are then converted into phrasal edits for producing DLFs in both directions, where the disjunctions represent alternative choices at the level of semantic dependencies. The resulting DLFs are fed into the OpenCCG realizer for n-best realization, using a pruning strategy that encourages lexical diversity. After merging, the approach yields an n-best list of paraphrases that contain grammatical alternatives to each original sentence, as well as paraphrases that mix and match content from the pair. A preliminary error analysis suggests that the approach could benefit from taking the word order in the original sentences into account. We conclude with a discussion of plans for future work, highlighting the method's potential use in enhancing automatic MT evaluation.