tech,36-12-J05-1003,bq |
ranking tasks
</term>
, for example ,
<term>
|
speech recognition
|
</term>
,
<term>
machine translation
</term>
|
#8972
Although the experiments in this article are on natural language parsing (NLP), the approach should be applicable to many other NLP problems which are naturally framed as ranking tasks, for example,speech recognition, machine translation, or natural language generation. |
tech,39-12-J05-1003,bq |
,
<term>
speech recognition
</term>
,
<term>
|
machine translation
|
</term>
, or
<term>
natural language generation
|
#8975
Although the experiments in this article are on natural language parsing (NLP), the approach should be applicable to many other NLP problems which are naturally framed as ranking tasks, for example, speech recognition,machine translation, or natural language generation. |
measure(ment),6-8-J05-1003,bq |
<term>
model
</term>
achieved 89.75 %
<term>
|
F-measure
|
</term>
, a 13 % relative decrease in
<term>
|
#8841
The new model achieved 89.75%F-measure, a 13% relative decrease in F-measure error over the baseline model’s score of 88.2%. |
other,23-7-J05-1003,bq |
evidence from an additional 500,000
<term>
|
features
|
</term>
over
<term>
parse trees
</term>
that
|
#8822
The method combined the log-likelihood under a baseline model (that of Collins [1999]) with evidence from an additional 500,000features over parse trees that were not included in the original model. |
other,10-4-J05-1003,bq |
approach
</term>
is that it allows a
<term>
|
tree
|
</term>
to be represented as an arbitrary
|
#8720
The strength of our approach is that it allows atree to be represented as an arbitrary set of features, without concerns about how these features interact or overlap and without the need to define a derivation or a generative model which takes these features into account. |
other,37-4-J05-1003,bq |
overlap and without the need to define a
<term>
|
derivation
|
</term>
or a
<term>
generative model
</term>
|
#8747
The strength of our approach is that it allows a tree to be represented as an arbitrary set of features, without concerns about how these features interact or overlap and without the need to define aderivation or a generative model which takes these features into account. |
tech,40-4-J05-1003,bq |
define a
<term>
derivation
</term>
or a
<term>
|
generative model
|
</term>
which takes these
<term>
features
</term>
|
#8750
The strength of our approach is that it allows a tree to be represented as an arbitrary set of features, without concerns about how these features interact or overlap and without the need to define a derivation or agenerative model which takes these features into account. |
model,7-7-J05-1003,bq |
<term>
log-likelihood
</term>
under a
<term>
|
baseline model
|
</term>
( that of
<term>
Collins [ 1999 ]
</term>
|
#8806
The method combined the log-likelihood under abaseline model (that of Collins [1999]) with evidence from an additional 500,000 features over parse trees that were not included in the original model. |
other,14-3-J05-1003,bq |
<term>
ranking
</term>
, using additional
<term>
|
features
|
</term>
of the
<term>
tree
</term>
as evidence
|
#8703
A second model then attempts to improve upon this initial ranking, using additionalfeatures of the tree as evidence. |
tech,30-12-J05-1003,bq |
</term>
which are naturally framed as
<term>
|
ranking tasks
|
</term>
, for example ,
<term>
speech recognition
|
#8966
Although the experiments in this article are on natural language parsing (NLP), the approach should be applicable to many other NLP problems which are naturally framed asranking tasks, for example, speech recognition, machine translation, or natural language generation. |
tech,2-2-J05-1003,bq |
probabilistic parser
</term>
. The base
<term>
|
parser
|
</term>
produces a set of
<term>
candidate
|
#8665
The baseparser produces a set of candidate parses for each input sentence, with associated probabilities that define an initial ranking of these parses. |
tech,11-1-J05-1003,bq |
which rerank the output of an existing
<term>
|
probabilistic parser
|
</term>
. The base
<term>
parser
</term>
produces
|
#8660
This article considers approaches which rerank the output of an existingprobabilistic parser. |
measure(ment),14-8-J05-1003,bq |
</term>
, a 13 % relative decrease in
<term>
|
F-measure
|
</term>
error over the
<term>
baseline model
|
#8849
The new model achieved 89.75% F-measure, a 13% relative decrease inF-measure error over the baseline model’s score of 88.2%. |
other,20-5-J05-1003,bq |
ranking problems
</term>
described in
<term>
|
Freund et al. ( 1998 )
|
</term>
. We apply the
<term>
boosting method
|
#8779
We introduce a new method for the reranking task, based on the boosting approach to ranking problems described inFreund et al. ( 1998 ). |
other,21-2-J05-1003,bq |
probabilities
</term>
that define an initial
<term>
|
ranking
|
</term>
of these
<term>
parses
</term>
. A second
|
#8684
The base parser produces a set of candidate parses for each input sentence, with associated probabilities that define an initialranking of these parses. |
other,10-3-J05-1003,bq |
attempts to improve upon this initial
<term>
|
ranking
|
</term>
, using additional
<term>
features
</term>
|
#8699
A second model then attempts to improve upon this initialranking, using additional features of the tree as evidence. |
other,12-2-J05-1003,bq |
candidate parses
</term>
for each input
<term>
|
sentence
|
</term>
, with associated
<term>
probabilities
|
#8675
The base parser produces a set of candidate parses for each inputsentence, with associated probabilities that define an initial ranking of these parses. |
tech,4-5-J05-1003,bq |
</term>
into account . We introduce a new
<term>
|
method
|
</term>
for the
<term>
reranking task
</term>
|
#8763
We introduce a newmethod for the reranking task, based on the boosting approach to ranking problems described in Freund et al. (1998). |
tech,6-9-J05-1003,bq |
The article also introduces a new
<term>
|
algorithm
|
</term>
for the
<term>
boosting approach
</term>
|
#8867
The article also introduces a newalgorithm for the boosting approach which takes advantage of the sparsity of the feature space in the parsing data. |
tech,2-8-J05-1003,bq |
original
<term>
model
</term>
. The new
<term>
|
model
|
</term>
achieved 89.75 %
<term>
F-measure
</term>
|
#8837
The newmodel achieved 89.75% F-measure, a 13% relative decrease in F-measure error over the baseline model’s score of 88.2%. |