Skip to content

headrule_format.md

Latest commit

 

History

History
17 lines (11 loc) · 1.59 KB

headrule_format.md

File metadata and controls

17 lines (11 loc) · 1.59 KB

Headrule Format

A headrule file contains head-finding rules for all phrases (or clauses) in constituent trees. Each line consists of a head-finding rule for a phrase type defined in the following format:

<tag>\t<direction=l|r>\t<rule>(;<rule>)*
  • <tag>: phrasal/clausal level constituent tag (e.g., S, VP, NP).
  • <direction>: l - find the leftmost constituent, r - find the rightmost constituent.
  • <rule>: constituent/function tag rules in Java regular expression format.

Here is a head-finding rule for a noun phrase (NP) extracted from our headrule file, headrule_en_stanford.txt.

NP  r  NN.*|NML;NX;PRP;FW;CD;NP;-NOM;QP|JJ.*|VB.*;ADJP;S;SBAR;.*
  • For a noun phrase, it first searches for the rightmost constituent whose tag is NN.*|NML, which covers {NN, NNS, NNP, NNPS, NML}. The rightmost constituent with any of these tags becomes the head of this noun phrase. If such a constituent does not exist, it searches for the rightmost constituent whose tag is NX. This procedure is repeated until the head is found. By default, the rightmost constituent with any tag (.*) becomes the head of this phrase.
  • - is used to indicate a function tag. After searching for a constituent whose tag is NP, it searches for any constituent with the function tag NOM.
  • See the Appendix A for more details about the constituent and function tags.