Fairhaven2022 Parsing WordNet definition and linking to senses

Discussion: Glosstag -- parsing wordnet definitions and linking to senses

Moderator: Alexandre

Scribe: Yi-Chien

Slides: glosstag.pdf

Minutes

Overview

• Nice to have translations for

• ERG Processing:

  • How to map the annotation? Mapping annotation to syntactic tree, MRS, or DMRS?
  • PSOS tagging mismatches
  • Tokenization mismatches with ERG (RPP)
  • Difference between versions -- which version to use?

• Problems in tokenization: extra/missing punctuations; How to deal with the punctuation formatting

• Multiword expressions: How to deal with multiword expressions:

  • Multiple words in proper nouns (e.g. San Diego)
  • MWT to single predicates (e.g. speed up)
  • NN compounds are not always MWE in WN
  • adj-nouns are sometimes MWE in WN

• Annotation:

  • Emacs
  • Deginitions and examples are parsed and mapped to MRS predicates to tokens/senses. Mismatches can be heuristically, trained or manually solved.
  • Make data available for browsing
  • Possible directions:
    • how can sense annotations help parsing selection?
    • continuous treebanking and sense annotation alignment?
  • version difference between PWN
  • ERG enhanced lexicon

• Ideas

Discussion

• Francis: Need to decide whether we want to tag all the multiword expressions. Add idiomatic to ERG. Tag compositional as single words.
• Eric: Can I take the MRS and get the possible links to the WN?
• Francis: Yes, will be a model to determine which one is the most likely one
• Luis: How many unique senses are actually unique? How many senses do we need to use since it might be keep growing?
• Eric: Is there a GUI for us to tell the that links to the MRS are the ones newly added?
• Francis: There is tools that tell link of the same sense between different version
• Alexandre: How to differentiate some senses that are highly similar
• Luis: We need to allow ambiguity and to select the core senses
• Alexandre: An IBM work proposed a method to map the WN to gloss (using some hash map).
• Guy: comment on the 1 pt. How are you going to determine where to draw the distinguishment. How much extra meaning to you want to get from compound nouns.
• Luis: Is we have one sense that matches one of the word in the word, we use that sense
• Francis: How to not tag "hot dog" that is "a dog that is hot"
• Eric: If we get the different representations in MRS, we can't have that representation so "red deer" can be interpreted as "a deer that is red" (mapping to the sense "red" and "deer" in WN). But "hot dog" cannot be interpreted as "a dog that is hot" (mapping to the sense "hot" and "dog" in WN).