Mining Diagnostic Text Reports by Learning to Annotate Knowledge Roles - Natural Language Processing and Text Mining - page 56

mation included in the output. Figure 4.9 shows the output of the same sentence 9
parsed by each parser, and in the following, we discuss each of them.

(ROOT
(NUR
(S
(PP (APPR Auf) (CARD NS))
(VAFIN wurden)
(VP
(AP (ADV ca.)
(NM (CARD 5)
(CARD gerissene)
(CARD Keilsicherungsbandagen)))
(VVPP festgestellt)))
($..)))

Stanford Parser

(utt:
(S.fin:
(PP: (APPR: Auf)
(NN: NS))
(VWFIN: wurden)
(AP: (AVP-MAD: (ADV-MAD: ca.))
(CARD: 5))
(NP.nom: (AP: (ADJA%: gerissene))
(NN.nom: Keilsicherungsbandagen))
(VVPP%: festgestellt)))
(\$.: .))

BitPar Parser

(TOP
(S
(PP-MO (APPR-AD Auf)
(NE-NK NS) )
(VAFIN-HD wurden)
(NP-SB
(ADV-MO ca.) (CARD-NK 5)
(ADJA-NK gerissene)
(NN-NK Keilsicherungsbandagen))
(VP-OC (VVPP-HD festgestellt)))
($. .))

Sleepy Parser

Auf NS wurden ca. 5 gerissene Keilsicherungsbandagen festgestellt.
On NS were ca. 5 torn wedge's safety bands detected.

Fig. 4.9. Parsing output of the same sentence from the three parsers

Stanford Parser - The Stanford parser is an ambitious project that tackles the task
of generating parse trees from unlabeled data independently of the language.
For the moment, the parser is distributed with parameter files for parsing En-
glish, German, and Chinese. We tested the parser on our data and noticed that

9 English translation: “On NS were detected circa 5 torn wedge's safety bands.”