Melvin's digital garden

POST in Prolog

CREATED: 200703270856 ** Background

  • part of speech tagging
  • transformation-based automatic POST (Brill 1995)
  • applications ** spelling/grammar checking ** speech synthesis ** knowledge extraction ** Approach
  • plain text, initial tags, rules -> iteratively select and apply best rule
  • best = greatest decrease in #errors of training text
  • types of rules ** replacement ** addition ** deletion ** reduction
  • performance measures ** Score = |P| - |N| ** Accuracy = |P| / (|P| + |N|)

** Probabilistic extension

  • keep all tags but with probabilities
  • update probabilities on rule match

Links to this note