cogcomp-nlp
cogcomp-nlp copied to clipboard
CogComp
Shallow parse: incorrectly tagging gerunds
Quoting from Jeff Schneider's email:
Not sure if you guys care about this level of detail, but the shallow parser (which I LOVE) is incorrectly tagging gerunds as part of the verb phrase.
"running caused a panic", "sleeping is my hobby", etc.
(I figured that our issue solving system is a "stack" and old issues are never touched, so I briefly browsed old issues and found this one.)
Sleeping is my hobby.
The panic was caused by running people.
Running caused a panic.
Running people caused a panic.
I think the POS tagging is correct: (VBG Sleeping) (VBZ is) (PRP$ my) (NN hobby) (. .) (DT The) (NN panic) (VBD was) (VBN caused) (IN by) (VBG running) (NNS people) (. .) (VBG Running) (VBN caused) (DT a) (NN panic) (. .) (VBG Running) (NNS people) (VBD caused) (DT a) (NN panic) (. .)
Shallow parsing has some mistakes as mentioned by Jeff: [VP Sleeping ] [VP is ] [NP my hobby ] This is correct I guess. [NP The panic ] [VP was caused ] [PP by ] [NP running people ] This is correct I guess. [VP Running caused ] [NP a panic ] This is wrong I guess. [VP Running ] [NP people ] [VP caused ] [NP a panic ] This is wrong I guess.
Anyone knows the ground truth of shallow parsing? How was it defined? Do we want this issue open?
Anyone knows the ground truth of shallow parsing? How was it defined? Do we want this issue open?
Not sure; but I see that in the there is a row for "chunking" ( https://docs.google.com/spreadsheets/d/13e27MJjQnFO2is0LGe5T-4N11cTnSy2JSrSqhQIaKMs/edit#gid=0 ). @mayhewsw might know more about this.
Hi @mayhewsw , do you have some insight here? My feeling is that this is something unavoidable in system predictions. If you agree with this, then I would suggest close this issue.