Write a CFG parser for the Penn Treebank using the
data in /u/cs{2,4}48/data/parse. The training phase should:
- Extract rules
- Binarize rules
- Count rules
The parser should:
- have a per-span beam
- handle unary productions
- handle unknown words
Use this script to evaluate your parser output:
/u/cs448/evalb/evalb gold-file test-file
Plot accuracy vs. time as you vary the beam size.