D08-1027 |
improvement by controlling for
|
labeler
|
bias . Acknowledgments Thanks
|
C04-1186 |
paper , a novel semantic role
|
labeler
|
based on dependency trees is
|
C96-1033 |
accessed when the Linguistic Feature
|
Labeler
|
indicates a lexical feature value
|
A00-2008 |
is added by an automatic phrase
|
labeler
|
developed by the technical team
|
D08-1027 |
across all sets of the five expert
|
labelers
|
( " NE vs. E " ) . We then calculate
|
D08-1056 |
cate - gory/answer pairs . Each
|
labeler
|
labeled every pair with one of
|
D08-1056 |
data , we asked three volunteer
|
labelers
|
to label 1000 total cate - gory/answer
|
C96-1033 |
sentence . The Linguistic Feature
|
Labeler
|
attaches features and atomic
|
D08-1027 |
. Our evaluation of non-expert
|
labeler
|
data vs. expert annotations for
|
A94-1013 |
an automatic sentence boundary
|
labeler
|
which uses probabilistic part-of-speech
|
C96-1033 |
passing it to the Linguistic Feature
|
Labeler
|
, which adds semantic labels
|
D08-1071 |
labeler and f2 , the " true " NER
|
labeler
|
. ( Note that we assume f1 E
|
D08-1032 |
used the ASSERT semantic role
|
labeler
|
system to parse the sentence
|
D08-1027 |
responses of the remaining five
|
labelers
|
on that set . In this way we
|
A94-1013 |
Results We tested the boundary
|
labeler
|
on a large body of text containing
|
C96-1001 |
annotation instructions used to train
|
labelers
|
to segment spoken discourses
|
C96-1033 |
ESST , the Linguistic Feature
|
Labeler
|
and Chunk Relation Finder networks
|
D08-1027 |
, and frequently within only 2
|
labelers
|
. Pooling judgments across all
|
D08-1027 |
and since multiple non-expert
|
labelers
|
may contribute to a single set
|
D08-1056 |
This was mostly left up to the
|
labelers
|
; the only suggestion was that
|