sentence boundary detection task
</term>
in
<term>
speech
</term>
, although it is interesting to note
#9558In general, our CRF model yields a lower error rate than the HMM and Maxent models on the NIST sentence boundary detection task inspeech, although it is interesting to note that the best results are achieved by three-way voting among the classifiers.
other,9-1-P05-1056,ak
speech
</term>
is important for enriching
<term>
speech recognition output
</term>
, making it easier for humans to
#9431Sentence boundary detection in speech is important for enrichingspeech recognition output, making it easier for humans to read and downstream modules to process.
other,19-4-P05-1056,ak
<term>
human transcriptions
</term>
and
<term>
speech recognition output
</term>
. In general , our
<term>
CRF model
#9529We evaluate across two corpora (conversational telephone speech and broadcast news speech) on both human transcriptions andspeech recognition output.
other,29-2-P05-1056,ak
knowledge sources
</term>
for detecting
<term>
sentence boundaries
</term>
. In this paper , we evaluate the
#9477In previous work, we have developed hidden Markov model (HMM) and maximum entropy (Maxent) classifiers that integrate textual and prosodic knowledge sources for detectingsentence boundaries.
tech,44-5-P05-1056,ak
<term>
three-way voting
</term>
among the
<term>
classifiers
</term>
. This probably occurs because each
#9577In general, our CRF model yields a lower error rate than the HMM and Maxent models on the NIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved by three-way voting among theclassifiers.
other,14-6-P05-1056,ak
strengths and weaknesses for modeling the
<term>
knowledge sources
</term>
. We present a framework for
<term>
#9593This probably occurs because each model has different strengths and weaknesses for modeling theknowledge sources.
lr,4-4-P05-1056,ak
prior work . We evaluate across two
<term>
corpora
</term>
( conversational telephone speech
#9514We evaluate across twocorpora (conversational telephone speech and broadcast news speech) on both human transcriptions and speech recognition output.
tech,40-5-P05-1056,ak
that the best results are achieved by
<term>
three-way voting
</term>
among the
<term>
classifiers
</term>
#9573In general, our CRF model yields a lower error rate than the HMM and Maxent models on the NIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved bythree-way voting among the classifiers.
other,16-4-P05-1056,ak
and broadcast news speech ) on both
<term>
human transcriptions
</term>
and
<term>
speech recognition output
#9526We evaluate across two corpora (conversational telephone speech and broadcast news speech) on bothhuman transcriptions and speech recognition output.
other,22-2-P05-1056,ak
) classifiers
</term>
that integrate
<term>
textual and prosodic knowledge sources
</term>
for detecting
<term>
sentence boundaries
#9470In previous work, we have developed hidden Markov model (HMM) and maximum entropy (Maxent) classifiers that integratetextual and prosodic knowledge sources for detecting sentence boundaries.
model,5-6-P05-1056,ak
This probably occurs because each
<term>
model
</term>
has different strengths and weaknesses
#9584This probably occurs because eachmodel has different strengths and weaknesses for modeling the knowledge sources.
other,19-5-P05-1056,ak
HMM and Maxent models
</term>
on the
<term>
NIST sentence boundary detection task
</term>
in
<term>
speech
</term>
, although it
#9552In general, our CRF model yields a lower error rate than the HMM and Maxent models on theNIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved by three-way voting among the classifiers.
tech,0-1-P05-1056,ak
5
<term>
ACE relation types
</term>
.
<term>
Sentence boundary detection
</term>
in
<term>
speech
</term>
is important
#9422Evaluation on the ACE corpus shows that effective incorporation of diverse features enables our system outperform previously best-reported systems on the 24 ACE relation subtypes and significantly outperforms tree kernel-based systems by over 20 in F-measure on the 5 ACE relation types.Sentence boundary detection in speech is important for enriching speech recognition output, making it easier for humans to read and downstream modules to process.
other,4-1-P05-1056,ak
Sentence boundary detection
</term>
in
<term>
speech
</term>
is important for enriching
<term>
speech
#9426Sentence boundary detection inspeech is important for enriching speech recognition output, making it easier for humans to read and downstream modules to process.
model,13-5-P05-1056,ak
lower
<term>
error rate
</term>
than the
<term>
HMM and Maxent models
</term>
on the
<term>
NIST sentence boundary
#9546In general, our CRF model yields a lower error rate than theHMM and Maxent models on the NIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved by three-way voting among the classifiers.
measure(ment),9-5-P05-1056,ak
<term>
CRF model
</term>
yields a lower
<term>
error rate
</term>
than the
<term>
HMM and Maxent models
#9542In general, our CRF model yields a lowererror rate than the HMM and Maxent models on the NIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved by three-way voting among the classifiers.
tech,7-2-P05-1056,ak
previous work , we have developed
<term>
hidden Markov model ( HMM ) and maximum entropy ( Maxent ) classifiers
</term>
that integrate
<term>
textual and prosodic
#9455In previous work, we have developedhidden Markov model ( HMM ) and maximum entropy ( Maxent ) classifiers that integrate textual and prosodic knowledge sources for detecting sentence boundaries.
tech,24-3-P05-1056,ak
task and relate results with this
<term>
model
</term>
to our prior work . We evaluate across
#9504In this paper, we evaluate the use of a conditional random field (CRF) for this task and relate results with thismodel to our prior work.
model,4-5-P05-1056,ak
recognition output
</term>
. In general , our
<term>
CRF model
</term>
yields a lower
<term>
error rate
</term>
#9537In general, ourCRF model yields a lower error rate than the HMM and Maxent models on the NIST sentence boundary detection task in speech, although it is interesting to note that the best results are achieved by three-way voting among the classifiers.