BarcelonaParaphrasing

Discussion: Paraphrasing

Moderator: Francis Bond; Scribe: Michael Goodman

Objective

• Share information on what is being done
  • parse and generate
    • problems with unknown words --- can't generate from MRS
    • can use Erik's generation model (in batch mode)
  • monolingual translation

    • EnEn

      • mainly underspecification of => poss_rel => compound_or_prep

        • minutes of meeting/meeting's minutes/meeting minutes/minutes of meeting's

      • need to build a scoring model for the underspecified MRS

      • need a full MT model for transfer rules

    • What about lexical? meeting/transactions --- massively larger TFS or external rules?

    • can everything be done by underspecification?

• Applications
  • SMT training data
  • query expansion
  • normalization
  • relaxed input for MT
  • debugging
  • anything else?
• How do we acquire paraphrase rules?
• Discuss ways to make paraphrasing more robust
  • generation often freezes (host lisp running out of memory?)
  • generation debugging tools

Notes

Some existing discussions:

• [wiki:RmrsParaphrasing Early discussions]

• [wiki:ParaTop Paraphrasing Top page]