learn.ifs.name

Coreference Resolution and Entity Linking

Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24

6 cards in “24.3 Mention detection” · whole deck

24.3 Mention detection

  1. How do mention detection algorithms usually propose candidates?
    Very liberally, favoring recall and filtering later: all NPs, possessive pronouns and named entities from parsers and NER, or (span-based systems) all n-gram spans up to about 10 words.
  2. How do rule-based systems detect pleonastic it?
    Patterns with dictionaries of modal adjectives and cognitive verbs (Lappin and Leass): 
    It is Modaladjective that S
It is Modaladjective (for NP) to VP
It is Cogv-ed that S
It seems/appears/means/follows (that) S
  3. Give the common rule-based first-pass mention detector for OntoNotes (Lee et al. 2013).
    (1) Take all NPs, possessive pronouns and named entities. (2) Remove numeric quantities, mentions embedded in larger mentions, adjectival nation forms and stop words. (3) Remove pleonastic it by regex patterns.
  4. What learned components help mention detection?
    Referentiality classifiers, anaphoricity classifiers (is this NP an anaphor?) and discourse-new classifiers (is it new and a potential antecedent?), using features like head word, context, definiteness, animacy, length and position.
  5. Why does hard filtering by an anaphoricity classifier work poorly, and what is used instead?
    A high threshold loses recall; a low one admits pleonastic mentions and hurts precision. Modern systems do mention detection, anaphoricity and coreference jointly in one end-to-end model.
  6. How did Bergsma et al. use web n-grams to detect non-anaphoric it?
    Turn the context into a pattern (make * in advance, make * in Hollywood) and see what words fill the wildcard: non-anaphoric (idiomatic) contexts mostly allow only it, anaphoric ones allow many NPs (make them in advance).

Source