nlp - Why there is a difference in parse tree output generated from api and GUI provided in stanfordNLP -
i using 'stanford-corenlp-full-2013-06-20' api generate parse tree given below
private string text= "heart attack causes reduced lifespan average"; annotation annotation = new annotation(text); corenlp.annotate(annotation); list<coremap> sentences = annotation.get(sentencesannotation.class); (coremap sentence : sentences) { tree tree = sentence.get(treeannotation.class); tree.pennprint(); }
it showing sub sentence 's' shown below
(root (**s** (np (nnp heart) (nn attack)) (vp (vbz causes) (**s** (np (vbn reduced) (nn lifespan) (nn average))))))
but when try parse same sentence using gui provided 'stanford-parser-full-2013-06-20' giving different tree (it seems right one) given below
(root (**s** (np (nnp heart) (nn attack)) (vp (vbz causes) (vp (vbn reduced) (np (nn lifespan) (nn average))))))
can 1 point out why both showing 2 different outputs though both belong same version.
the stanford parser output different results depending on number of annotation tasks asking (source). required parser output sentence split, tokenization, , parse tasks. however, if run sentence spilt, tokenization, part-of-speech tag, , parse tasks different results.
so corenlp annotation going add pos tagging default, giving different parse results parse task.
in experience working parse trees , both forms of output neither method strictly better.
Comments
Post a Comment