Chapter 21
Information Extraction: Relations, Events, and Time
Chapter: Information Extraction: Relations, Events, and Time, 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.
- Relation extraction p. 2
- Pattern matching p. 4
- Supervised classifiers p. 5
- Bootstrapping p. 7
- Distant supervision p. 8
- Unsupervised extraction p. 10
- How relations are scored p. 12
- Events and time p. 12
- Event extraction p. 12
- Time representation and Reichenbach's reference point p. 14
- Aspect p. 16
- TimeBank p. 17
- Temporal expressions, normalization, and ordering p. 18
- Template filling p. 22
- Filling a fixed schema p. 22
- Machine-learning fillers p. 22
- Earlier finite-state systems p. 23
FAQ
Does distant supervision mean the relation label was annotated in that sentence?
No. It projects a known relation between two entities onto sentences that mention both. The sentence may not express the relation. That noise is part of the method.
What is temporal normalization?
Mapping an expression such as a relative date onto a specific time value. Extracting the phrase and resolving what time it names are different steps.
Why is a reference point needed for tense?
Event time and speech time are not enough for every tense. Reichenbach's reference point is the extra time the chapter uses to describe how tense is anchored.
Can a filled template be correct if one slot has the wrong entity?
The template is only as good as its slots. A score has to say whether it requires every slot, or scores slots separately. One right slot does not make the record right.