tech,4-1-H05-1012,ak phrases </term> . This paper presents a <term> maximum entropy word alignment algorithm </term> for <term> Arabic-English </term> based
other,10-1-H05-1012,ak word alignment algorithm </term> for <term> Arabic-English </term> based on <term> supervised training
lr,13-1-H05-1012,ak <term> Arabic-English </term> based on <term> supervised training data </term> . We demonstrate that it is feasible
lr,8-2-H05-1012,ak demonstrate that it is feasible to create <term> training material </term> for problems in <term> machine translation
tech,13-2-H05-1012,ak training material </term> for problems in <term> machine translation </term> and that a mixture of <term> supervised
tech,20-2-H05-1012,ak translation </term> and that a mixture of <term> supervised and unsupervised methods </term> yields superior performance . The
model,1-3-H05-1012,ak </term> yields superior performance . The <term> probabilistic model </term> used in the <term> alignment </term>
tech,6-3-H05-1012,ak probabilistic model </term> used in the <term> alignment </term> directly models the link decisions
tech,3-4-H05-1012,ak decisions . Significant improvement over <term> traditional word alignment techniques </term> is shown as well as improvement on
measure(ment),15-4-H05-1012,ak as well as improvement on several <term> machine translation tests </term> . Performance of the <term> algorithm
tech,3-5-H05-1012,ak translation tests </term> . Performance of the <term> algorithm </term> is contrasted with <term> human annotation
measure(ment),7-5-H05-1012,ak algorithm </term> is contrasted with <term> human annotation performance </term> . The paper presents a <term> Bayesian
skrij podrobnosti