Accurate Unlexicalized Parsing

1 Experimental Setup

1. What was Klein and Manning’s experimental setup?
   Train on WSJ sections 2–21, develop on 393 sentences of section 22 (a partly manual hill-climb on features), test on section 23. Trees are transformed, then rule probabilities are unsmoothed maximum-likelihood estimates, parsed exhaustively with a generalized CKY parser.
   • Klein & Manning, 2003, Accurate Unlexicalized Parsing (Proceedings of the 41st Annual Meeting of ACL), §1, p. 424
2. How did the unlexicalized parser handle unknown words?
   Only tagging probabilities were smoothed: words map to classes by capitalization, suffix, digits and other character features, and \(P(\text{tag} \mid \text{word}) = \frac{c(\text{tag}, \text{word}) + \kappa P(\text{tag} \mid \text{wordclass})}{c(\text{word}) + \kappa}\), then inverted to \(P(\text{word} \mid \text{tag})\).
   • Klein & Manning, 2003, Accurate Unlexicalized Parsing (Proceedings of the 41st Annual Meeting of ACL), §1, p. 424