PCFG object beta complete

https://github.com/aetilley/pcfg

Object has both a parser and a scorer.

Proposed future strategy:

Use penn treebank or some large public treebank to train generic PCFG *parser*. In particular, need to get a counts file like

https://github.com/usami/pcfg/blob/master/counts_file.sample

Use our trained parser to parse WP revisions (regular and vandalous) in order to get *two more counts* files to train two PCFG *scorers* p_vandal and p_regular. Looks like the tokenizer will be straightforward (thanks Aaron:
https://gist.github.com/halfak/1620beae124716504cba)

Add features, for, say

min_{s \in revision}(log(p_{vandal}(s))) - min_{s \in revision}(log(p_{regular}(s)))