Coreference Resolution and Entity Linking

24.4 Architectures for coreference algorithms

1. Along which two dimensions does Ng (2010) categorize coreference architectures?
   Entity-based (representing discourse entities) vs mention-based (considering mentions independently), and whether ranking models directly compare candidate antecedents.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 551
2. What is the mention-pair architecture?
   A binary classifier given a candidate anaphor and a candidate antecedent decides coreferring or not, e.g. \(p(coref \mid \text{Victoria Chen}, \text{she})\) should be high and \(p(coref \mid \text{Megabucks Banking}, \text{she})\) low.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 551
3. How are training pairs selected for the mention-pair model (Soon et al. 2001)?
   For each anaphor: one positive pair with its closest antecedent, and a negative pair for each mention in between. For the company: positive (the company, Megabucks Banking); negatives (the company, her), (the company, her pay), (the company, the 38-year-old).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, pp. 551–552
4. What are closest-first and best-first clustering?
   After pairwise classification: closest-first runs right to left from mention \(i-1\) and links the first antecedent with probability > .5; best-first links the most probable antecedent. Then take the transitive closure.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 552
5. What are the two main problems of the mention-pair model?
   It doesn’t compare candidate antecedents with each other, and it ignores the discourse model: each decision is local to a pair, without other mentions of the same entity.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 552
6. What is the mention-ranking architecture?
   For mention \(i\), a softmax over all prior mentions plus a dummy \(\epsilon\) (no antecedent: discourse-new or non-anaphoric), jointly learning anaphoricity and coreference.

   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 552
7. Why is training trickier for mention-ranking than for mention-pair models?
   The best antecedent is latent: the gold clustering gives a whole set of legal antecedents. Heuristics pick one (e.g. the closest), or the loss sums the probability of all correct antecedents.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 553
8. What are entity-based coreference models?
   They link a mention to a previous discourse entity (cluster) rather than a mention, using cluster features (size, shape like P-Pr-D for proper–pronoun–definite, aggregated pair probabilities) or learned cluster representations. Gains have been small, so mention ranking is more common.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 24, §24.4, p. 553