Parser output on shell is different from the demo output

[paridhij]

I tried a sentence "DES reduces the risk of recurrent MI, CV death and all-cause mortality" on the TurboParser online demo and the downloaded TurboParser2.3.0. I am using parse.sh to get the conll format of the parse tree. Here are the outputs:

Demo output [which is correct]: 1 DES NN NNP 2 nsubj 2 reduces VB VBZ 0 null 3 the DT DT 4 det 4 risk NN NN 2 dobj 5 of IN IN 4 prep 6 recurrent JJ JJ 7 amod 7 MI NN NNP 5 pobj 8 , , , 7 punct 9 CV NN NNP 10 nn 10 death NN NN 7 conj 11 and CC CC 7 cc 12 all-cause JJ JJ 13 amod 13 mortality NN NN 7 conj

while the TurboParser2.3.0 installed version gives:

1 DES NNP NNP 2 SUB 2 reduces VBZ VBZ 0 ROOT 3 the DT DT 4 NMOD 4 risk NN NN 2 OBJ 5 of IN IN 4 NMOD 6 recurrent JJ JJ 7 NMOD 7 MI NNP NNP 13 NMOD 8 , , , 13 P 9 CV NNP NNP 13 NMOD 10 death NN NN 13 NMOD 11 and CC CC 13 NMOD 12 all-cause NN NN 13 NMOD 13 mortality NN NN 5 PMOD

Lines 8,9,10,11 have wrong dependencies, and node 13 has incorrect head. What am I missing here.

[paridhij]

I also tried with all the tree models, basic, full and standard. None of the models give the results demo gives.

[paridhij]

Can you please look into the discrepancy?

[andre-martins]

Hi,

Which demo are you refering to? http://demo.ark.cs.cmu.edu/parse?

The reason for the discrepancy is that you are using two different models. Note that the dependency labels are different (Stanford dependencies in the first example, in the demo model; Yamada-Matsumoto dependencies in the second one, your local model). To get the same dependencies as the demo, please contact Lingpeng Kong (CC'ed) who may be able to point you to the model being used at the demo.

Thanks,

André

2016-03-17 6:06 GMT+00:00 paridhij notifications@github.com:

I tried a sentence "DES reduces the risk of recurrent MI, CV death and all-cause mortality" on the TurboParser online demo and the downloaded TurboParser2.3.0. I am using parse.sh to get the conll format of the parse tree. Here are the outputs:

Demo output [which is correct]: 1 DES NN NNP 2 nsubj 2 reduces VB VBZ 0 null 3 the DT DT 4 det 4 risk NN NN 2 dobj 5 of IN IN 4 prep 6 recurrent JJ JJ 7 amod 7 MI NN NNP 5 pobj 8 , , , 7 punct 9

CV NN NNP 10 nn 10 death NN NN 7 conj 11 and CC CC 7 cc _ 12 all-cause JJ JJ 13 amod 13 mortality NN NN 7 conj _

while the TurboParser2.3.0 installed version gives:

1 DES NNP NNP 2 SUB 2 reduces VBZ VBZ 0 ROOT 3 the DT DT 4 NMOD 4 risk NN NN 2 OBJ 5 of IN IN 4 NMOD 6 recurrent JJ JJ 7 NMOD 7 MI NNP NNP 13 NMOD 8 , , , 13 P 9 CV NNP NNP 13 NMOD 10 death NN NN 13 NMOD 11 and CC CC 13 NMOD 12 all-cause NN NN 13 NMOD 13 mortality NN NN 5 PMOD

Lines 8,9,10,11 have wrong dependencies, and node 13 has incorrect head. What am I missing here.

— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/andre-martins/TurboParser/issues/24

[paridhij]

Thanks Andre.

[LittleYUYU]

Is the problem solved? Actually I also want the Turbo parser being used in the demo. Could you also put this version online? Thanks much!

(Update: Already got the demo version. Thx!)

[qili5]

I am also very interested to get the Turbo parser used in the demo. http://demo.ark.cs.cmu.edu/parse This is so far the most accurate parser for my corpus.