Compromise: syntactic parsing

[nelson-liu]

Hi,
In the readme, you state that you get a "86% on the Penn treebank". Are you doing parsing in this library? If so, I wasn't able to find any docs on it, could you perhaps guide me to the correct module. If not, what do you mean when you say "86% on the Penn treebank"?

Thanks!

[spencermountain]

hi nelson! yeah, i haven't done a good job at keeping these organised. for the penn version, i used
this file (big) and mapped its tagset to mine, which got messier and messier. Importantly, I ignored sentences that had different # of terms, and only gave a yes/no if the sentence proved to be a perfect match.

This was about a year ago. Since I added phrasal-verb chunking, and a bunch of adhoc lumping-term rules, it became harder and harder to compare the two datasets.

i've kept the other performance tests in ./tests/perf_tests, if you wanna run them. I should probably figure-out a way to properly evaluate POS again, now that it's strayed from the Penn output considerably.

i do some sanity test stuff on a wikipedia corpus in ./tests/perf_tests/pos, but it doesn't give a % correct result, only which methods are heating up, etc.
cheers

[nelson-liu]

Ah ok, I see. Thanks for the detailed explanation! This library looks great :)

[spencermountain]

btw, this is incredible.

you're a legend.
cheers