Tags:	[CRF]

Shallow Parsing with Conditional Random Fields

Fei Sha & Pereira 2003

This paper describes an implementation of a chunker using IOB encoding with a CRF.

The following table lists the features used in the chunker:

Feature	Description
	The current output tag.
	The current state (current and previous output tag).
	A tuple of the current state and the previous state (current and previous two output tags).
	A tuple of the current state and the th word, .
	A tuple of the current state, the current word, and the previous word.
	A tuple of the current state, the current word, and the next word.
	A tuple of the current state and the part of speech tag of the th word, .
	A tuple of the current state and the two consecutive part of speech tags starting at word , .
	A tuple of the current state, and three consecutive part of speech tags centered on word , .
	A tuple of the current output tag and the th word, .
	A tuple of the current output tag, the current word, and the previous word.
	A tuple of the current output tag, the current word, and the next word.
	A tuple of the current output tag and the part of speech tag of the th word, .
	A tuple of the current output tag and the two consecutive part of speech tags starting at word , .
	A tuple of the current output tag, and three consecutive part of speech tags centered on word , .