Chapter 20
Dependency Parsing
Chapter: Dependency 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.
- Dependency relations p. 2
- Head and dependent p. 2
- Formalisms p. 3
- Projectivity p. 4
- Dependency treebanks p. 5
- Transition-based parsing p. 6
- Actions that build a tree p. 6
- An oracle from a gold tree p. 9
- Feature-based and neural classifiers p. 11
- Further transition systems p. 13
- Graph-based parsing p. 15
- Maximum spanning tree p. 16
- Feature-based arc scores p. 18
- Neural arc scores p. 19
- Evaluation p. 21
- Which predicted arcs count as correct p. 21
FAQ
What is a projective dependency tree?
A tree whose arcs do not cross when the words stay in sentence order. Some languages and some analyses need non-projective arcs. A parser that only builds projective trees cannot return those analyses.
What is an oracle in transition-based parsing?
A procedure that, given a gold tree, says which action the parser should have taken in a configuration. It creates training labels for the action classifier. It is not available at test time.
How does graph-based parsing differ from left-to-right actions?
It scores candidate arcs and searches for a tree, classically a maximum spanning tree. It does not have to commit to an early action before later words are scored.
Does a correct unlabeled head mean the parse is right?
Not if the evaluation includes labels. An arc can attach the right head and still have the wrong relation. Check which score the paper or chapter is reporting.