jgung / verbnet-parser Goto Github PK

What algorithms and heuristics are used to map srl output to verbnet frames. Also what dataset is used to train various models

semparse-core should also depend on clearwsd-api. VerbNetSenseClassifier.java refers to io.github.clearwsd.ParsingSensePredictor. This bug is in the master branch as of today, 4/2/20.

Add a Dockerfile for running semparse-web, instructions.

Hello again,

Another quick question: if I understood correctly the models were trained with the maximum length of 256 characters, so VerbNet Parser can't handle sequences with more than 256 characters. Could you please confirm my understanding?

Thank you very much for your help!

Thank you very much for sharing the parser. I wonder how the parse treats pb2vn mapping where pb predicate does not specific enough to identify the VN class?

I noticed that VN is more fine-grained for many cases and there seems to be critical differences in the results depending on. For instance "take.10" in prop bank can be mapped onto three VN classes "54.2", "103", or "54.3". Did you train the model with a VN specific annotated dataset to disambiguate this? OR did you manually chose one out of these three as default cases (not considering the other two?)?

If you took a statistical approach (the former), could you let me know where you got the VN annotated treebank?
If the latter is the approach, I wonder if you could share the overall strategy that you took.

I appreciate the way you handled this disambiguation task.
Thank you in advance!

Hello,

Thank you for the great repo!

I failed to process the following sentence using the docker image locally (and also on the website demo):
He was only a minute or so inside , and then I heard the key turn once more and he passed me and re - entered the house . Apparently something is wrong with the sequence re -: when deleting spaces or removing re, it works.

The error log: verbnet-parser-log-error.txt

Thank you for looking into that!