search

cslu-nlp / PerceptronixPointNever

Averaged perceptron part-of-speech tagger (à la Collins '02)
10 stars 2 forks source link

NP-chunking #3

Open kylebgorman opened 10 years ago

kylebgorman commented 10 years ago

Implement NP-chunking using the Ratnaparkhi/Collins feature set. This should ideally run in two modes:

  1. pre-computed tag mode (we're chunking treating those tags as gold)
  2. tag-and-chunk mode (we're going to tag, then reify those tags and use them for chunking)
kylebgorman commented 10 years ago

the thing that will help with #1 is to redo, to abstract, the PPN class so that it does not care what function(s) are used for extracting emission features. for #2, that should be something that the __main__ method handles, probably via a temp file.

kylebgorman commented 10 years ago

possible architecture for this: a PPN instance should be an instance variable of either Tagger or Chunker