search

microth / PathLSTM

Neural SRL model
71 stars 14 forks source link

Minor: upgrading stanford dependency would break anything? #12

Closed danyaljj closed 7 years ago

danyaljj commented 7 years ago

Would anything break if I use version 3.7.0 for Stanford dependency?

microth commented 7 years ago

What dependencies does the CogComp depparse model use? That's the one I'll use to retrain the SRL model. Everything else will lakely make things break (at least a little bit). //cc @christos-c

danyaljj commented 7 years ago

My question was with regards to your current "pre-illinois-built" branch, which uses corenlp parsing (3.6.0). Have you tried using 3.7.0? (the latest version). (Based on your previous response, I guess the answer is no)

microth commented 7 years ago

The "pre-illinois-built" branch uses the mate preprocessing pipeline (anna-3.x dependency), which is trained on CoNLL-2009 dependencies. The CoreNLP component is only used for tokenization.

microth commented 7 years ago

(As far as I remember, @christos-c added CoreNLP 3.6.0 to the pom file. Not sure whether this would have been necessary. Only Stanford's WSJTokenizer is needed--but maybe it was not available as a separate dependency?)

danyaljj commented 7 years ago

I see. Looks like nothing breaks. Cool.

christos-c commented 7 years ago

Sorry for not responding to this sooner, but yes, the main reason I added the dependency was the tokenizer. The new dep parser doesn't need anything from Stanford.