#4692Thelanguage model is initially estimated from a small manually segmented corpus of about 110,000 words.
lr,8-4-P03-1051,ak
</term>
is initially estimated from a
<term>
small manually segmented corpus
</term>
of about 110,000
<term>
words
</term>
#4699The language model is initially estimated from asmall manually segmented corpus of about 110,000 words.
other,30-7-P03-1051,ak
manually segmented corpus
</term>
of the
<term>
language
</term>
of interest . A central problem of
#4797We believe this is a state-of-the-art performance and the algorithm can be used for many highly inflected languages provided that one can create a small manually segmented corpus of thelanguage of interest.
lr,24-7-P03-1051,ak
</term>
provided that one can create a
<term>
small manually segmented corpus
</term>
of the
<term>
language
</term>
of interest
#4791We believe this is a state-of-the-art performance and the algorithm can be used for many highly inflected languages provided that one can create asmall manually segmented corpus of the language of interest.
measure(ment),10-6-P03-1051,ak
system
</term>
achieves around 97 %
<term>
exact match accuracy
</term>
on a
<term>
test corpus
</term>
containing
#4755The resulting Arabic word segmentation system achieves around 97%exact match accuracy on a test corpus containing 28,449 word tokens.
other,20-1-P03-1051,ak
sequence of
<term>
morphemes
</term>
in the
<term>
pattern
</term>
prefix * - stem-suffix * ( * denotes
#4622We approximate Arabic's rich morphology by a model that a word consists of a sequence of morphemes in thepattern prefix*-stem-suffix* (* denotes zero or more occurrences of a morpheme).
other,15-7-P03-1051,ak
algorithm
</term>
can be used for many
<term>
highly inflected languages
</term>
provided that one can create a
<term>
#4782We believe this is a state-of-the-art performance and the algorithm can be used for manyhighly inflected languages provided that one can create a small manually segmented corpus of the language of interest.
tech,17-2-P03-1051,ak
</term>
and uses it to bootstrap an
<term>
unsupervised algorithm
</term>
to build the
<term>
Arabic word segmenter
#4657Our method is seeded by a small manually segmented Arabic corpus and uses it to bootstrap anunsupervised algorithm to build the Arabic word segmenter from a large unsegmented Arabic corpus.
model,4-3-P03-1051,ak
</term>
. The
<term>
algorithm
</term>
uses a
<term>
trigram language model
</term>
to determine the most probable
<term>
#4676The algorithm uses atrigram language model to determine the most probable morpheme sequence for a given input.
tech,1-3-P03-1051,ak
unsegmented Arabic corpus
</term>
. The
<term>
algorithm
</term>
uses a
<term>
trigram language model
#4673Thealgorithm uses a trigram language model to determine the most probable morpheme sequence for a given input.
model,27-5-P03-1051,ak
corpus
</term>
, and re-estimate the
<term>
model parameters
</term>
with the expanded
<term>
vocabulary
#4735To improve the segmentation accuracy, we use an unsupervised algorithm for automatically acquiring new stems from a 155 million word unsegmented corpus, and re-estimate themodel parameters with the expanded vocabulary and training corpus.