Context-Based Machine Translation
http://groups.google.com/group/kunlp/browse_thread/thread/c33f13346ffeb43a/50322ccc03083148?lnk=raot
Context-Based Machine Translation
www.mt-archive.info/AMTA-2006-Carbonell.pdf
Spanish-to-English CBMT was tested on Spanish newswire text, achieving a
BLEU score of 0.6462 in June 2006, the highest BLEU reported for any
language pair.
Context-Based Machine Translation: A Disruptive Technology for
Very-High-Quality Automated Translation
——————————
*
Develop an efficient and practical technology, using novel algorithms, for
high-quality, context-based machine translation between languages.
* Sponsor: Meaningful Machines, LLC 1450 Broadway, 40th Floor
New York, NY 10018
- Project Performance Period: 11/1/2007 - 10/31/2009
- Total project (est.): $2,927,100.00
- Requested ATP funds: $2,000,000.00
After half a century of research, the vision of a language translation
machine-a computer capable of translating one natural human language into
another with a facility approaching human translators-remains a distant
prospect. Current machine translation (MT) systems fall into two categories:
rule-based and statistical. Rule-based systems rely on an extensive set of
manually-coded grammar and transfer rules between the source and target
languages developed by linguists and computer scientists. Statistical MT
emphasizes computer “learning” that derives translations from the analysis
of large blocks of pre-translated and organized parallel texts. Meaningful
Machines has proposed developing a third approach, Context-Based Machine
Translation (CBMT), based on new language-processing algorithms that use a
bilingual dictionary and a large body of text in the target language to
build and connect target language word strings in a manner that preserves
the context of the source language. CBMT, if practical, could easily be
extended to many different language pairs because it obviates the need to
hand-code extensive translation rules or to have access to large quantities
of parallel text. Meaningful Machines proposes to generalize their existing
system architecture to enable any-language to any-language translation, and
improve quality for the most challenging translation tasks, such as Chinese
to English. High-quality machine translation is potentially a $10 billion to
$30 billion market that will benefit the nation’s economy by facilitating
exports, addressing critical shortages of translators and opening up new
markets in areas such as daily newspapers, on-line chat services,
advertising, on-line searches and automated email translation.
–
Ergun Bicici
Koc University