Abstract:
Over the last seven years, we have witnessed significant progress
in the statistical approach to machine translation.
This progress has been achieved for both spoken
and written language in national and international projects.
In comparative evaluations for spoken language translation,
the statistical approach was found to be significantly superior
to the existing conventional approaches.
The first half of this talk will introduce the main components
of a statistical machine translation system (such as alignment and
lexicon models, generation of the target sentence)
and summarize achievements to date, with particular emphasis
on the author's experience in European projects.
The second part of the talk will be devoted to a discussion
of some important technical challenges and open research issues
in statistical machine translation.
Examples are the question of the correct form of Bayes decision rule,
the use of grammars and morphosyntax and, for spoken language,
the integration of recognition and translation.
Biography:
Hermann Ney has been working in the field of speech recognition,
natural language processing, and statistical modeling for
25 years and has authored and co-authored more than 200 papers
in international journals, conferences and books.
He is on the editorial board of several major scientific journals.
Since 1993, he has been a full professor of computer science at
RWTH Aachen (University of Technology) in Germany.
His work is motivated by the belief that the problem of
statistical modelling along with all its aspects
such as learning and decision making is the gateway to building
successful systems for speech and language processing.
