About the author: Jeff Allen has over a decade of experience in: translation, technical writing, localization, controlled language authoring software and writing guidelines, machine translation (MT) systems, translation memory (TM) tools, multilingual text and speech databases, speech recognition and synthesis technologies, OCR, SGML, XML, documentation Workflows, minority language engineering tools, language teaching, datawarehousing, and business intelligence reporting tools.
This article describes a project which specifically aimed at improving the quality of Machine
Translation (MT) output by creating:
a custom user Machine Translation (MT) dictionary, and
a set of preserved/Not-to-translate words, using a proven methodology by a well-trained MT software user.
The guiding principle is that it is possible to override general
dictionary entries and grammatical rules of the translation software in
order to create a very usable, good quality output text for further MT
postediting. The upfront text analysis and dictionary development can
be conducted in a short period of time, even conducted manually without
the help of (semi-)automated terminology extraction tools. This
approach to dictionary building also allows for a very optimal number
of dictionary entries (1/3 of the initially proposed candidates) to
produce the same translation as would take place with a full, maximal
set of dictionary entries.
Kodu
Click here to read full article in PDF format.
