Constraint-Based Part-of-Speech Tagging

EasyChair Preprint 8379

Abstract

This paper describes a constraint-based part-of-speech (POS) tagger, named CPOST, which treats POS tagging as a constraint satisfaction problem (CSP). CPOST treats each word as a variable, uses a lexicon to determine the domains of variables, employs context constraints to reduce ambiguity, and utilizes statistical models to label variables with values. This paper shows that, with a small number of context constraints that encode some of the basic linguistic knowledge, CPOST significantly enhances the precision at identifying base-form verbs, and mitigates the burden on syntax parsing.

Keyphrases: NLP, POS tagging, constraints

@booklet{EasyChair:8379,
  author    = {Neng-Fa Zhou},
  title     = {Constraint-Based Part-of-Speech Tagging},
  howpublished = {EasyChair Preprint 8379},
  year      = {EasyChair, 2022}}