Possible target corpora

NB: Projects do not have to be limited to this list of potential target corpora. Any English corpus that has semantic annotations over at least 100 sentences where the annotations do not come from the ERG is potentially fine (but please check with me).

• Groningen Meaning Bank
• ISI Abstract Meaning Representations (files in CommonView)
• Hyperproof text (files in CommonView)
• Zettlemoyer's LFs over Geoquery Geoquery (cf. Zettlemoyer and Collins 2005, files on CommonView)
• Zettlemoyer's LFs over ATIS (cf. Zettlemoyer and Collins 2007), files on CommonView)
• CLang
• QALD
• MASC, including PropBank, FrameNet, coreference annotations
• OntoNotes

Term project specs

• Choose target representation (with associated annotated corpus).
• Parse representative sample of the corpus with the ERG (at least 100 items, ideally more)
• Develop translation procedure to map ERG-produced MRSs to target representations. (Can be rule-based or stochastic.)
• Measure:
  • How much information in target representations is not mappable from ERG/MRS
  • How much information in MRS is suppressed (not mapped)
• Analyze (by looking at the data in detail):
  • What kinds of information are carried across
  • What kinds of information are mis-matched

Write up

• Describe target corpus (see corpus assignment)
• Describe mapping procedure
• Report on measurements of mappability (see evaluation plan)
• Report analysis of what is and isn't mapped
• Reflect on what the mismatches mean about semantics and meaning

Turn in the paper (as pdf) to CollectIt under "Term project", by 6/13.