AMTA 2002: From Research to Real Users Ever since the showdown between Empiricists and Rationalists a decade ago at TMI 92, MT researchers have hotly pursued promising paradigms for MT, including da- driven approaches (e.g., statistical, example-based) and hybrids that integrate these with more traditional rule-based components. During the same period, commercial MT systems with standard transfer archit- tures have evolved along a parallel and almost unrelated track, increasing their cov- age (primarily through manual update of their lexicons, we assume) and achieving much broader acceptance and usage, principally through the medium of the Internet. Webpage translators have become commonplace; a number of online translation s- vices have appeared, including in their offerings both raw and postedited MT; and large corporations have been turning increasingly to MT to address the exigencies of global communication. Still, the output of the transfer-based systems employed in this expansion represents but a small drop in the ever-growing translation marketplace bucket.At the probability threshold of 0.5, for Manual 1 the precision error was 5.6 times lower and the recall error was 8.0 times lower ... These substantial deletions stress the search strategy as well as the alignment models, since they force the initialanbsp;...
|Title||:||Machine Translation: From Research to Real Users|
|Author||:||Stephen D. Richardson|
|Publisher||:||Springer - 2003-06-30|