Appendix E
Statistical Constituency Parsing
Appendix: Statistical Constituency Parsing, in Daniel Jurafsky and James H. Martin, Speech and Language Processing, 3rd edition draft, August 19, 2026.
Topics to learn
Page numbers refer to the linked chapter PDF.
- Probabilistic context-free grammars p. 1
- Rule probabilities p. 1
- Disambiguation p. 2
- Language-model use p. 5
- Probabilistic CKY p. 6
- Estimating rule probabilities p. 7
- Where a plain PCFG fails p. 8
- Missed dependencies between rules p. 9
- No lexical sensitivity p. 9
- Stronger grammars p. 11
- Splitting nonterminals p. 11
- Lexicalized probabilistic CFGs p. 13
- The Collins parser p. 15
FAQ
What does a PCFG add to a context-free grammar?
A probability for each rule. The grammar can then prefer one licensed tree over another. It does not remove ambiguity by itself.
Why is lexical insensitivity a problem?
The same rule gets the same probability regardless of the words. Attachment preferences often depend on the head words. A plain PCFG cannot express that.
What is nonterminal splitting for?
It creates finer symbols so contexts that a coarse label treated as the same can have different rule probabilities. The split is a modeling choice, not a new linguistic category.
How is this appendix related to Chapter 19?
Chapter 19 introduces constituency, CKY, and span-based neural parsing. This appendix is the probabilistic and lexicalized parser line.