Integrating Translations from Multiple Sources within the PANGLOSS Mark III Machine Translation System

INTEGRATING TRANSLATIONS FROM MULTIPLE SOURCES WITHIN THE PANGLOSS MARK III MACHINE TRANSLATION SYSTEM Robert Frederking1, Sergei Nirenburg2, David Farwell2, Stephen Helmreich2, Eduard Hovy3, Kevin Knight3, Stephen Beale3, Constantine Domashnev1, Donalee Attardo1, Dean Grannes1, Ralf Brown1 1) Center for Machine Translation 2) Computing Research Laboratory Carnegie Mellon University New Mexico State University Schenley Park Box 30001 / 3CRL Pittsburgh, PA 15213-3890 Las Cruces, NM 88003 3) Information Sciences Institute of the University of S. California 4676 Admiralty Way Marina del Rey, CA 90292 Abstract Since MT systems, whatever translation method they employ, do not reach an optimum output on free text; each method handles some problems better than others. The PANGLOSS Mark III system is an MT environment that uses the best results from a variety of independent MT systems or engines working simultaneously within a single framework on the same text. This paper describes the method used to combine the outputs of the engines into a single text. 1 Introduction PANGLOSS Mark III is a multi-engine Spanish-to-English MT system. The PANGLOSS team is distributed at three sites — the Computing Research Laboratory of New Mexico State University, the Information Sciences Institute of the University of Southern California and the Center for Machine Translation of Carnegie Mellon University. Originally, PANGLOSS was supposed to be a pure knowledge-based machine translation (KBMT) system implemented in a version of the interlingua architecture. The project, however, evolved toward a more eclectic approach, mostly due to the necessity to perform well during periodical external evaluations whose timing and frequency was established after the project started. As KBMT systems rely on extensive knowledge bases, their gestation periods are typically longer than those of other kinds of systems. The first system configuration, the PANGLOSS Mark I system, was pure KBMT and did not perform well on the first evaluation in Summer 1992. The project team then decided to channel some of the resources into developing an interim evaluation system which would show an immediate improvement in output quality, while in parallel continuing to develop the "mainline" KBMT system. The idea was that at a certain stage the KBMT system would supplant the interim system. Thus, PANGLOSS Mark II was a simple lexical transfer system based on phrasal bilingual glossaries and machine- readable dictionaries. The evaluation results were better than during the first evaluation (White and O'Connell, 1994). 73 As our understanding of the MT task continued to evolve, it was decided not to discard the "interim" system in favor of the "mainline" one but rather to make these systems coexist and even include additional MT systems. In fact, our current system, PANGLOSS Mark III, differs from other MT systems because it employs not a single translation engine but a set of several engines whose results are integrated for the best overall output. PANGLOSS Mark III contains three distinct MT engines: • a KBMT system, the mainline Pangloss engine; • an example-based MT (EBMT) system (Nirenburg et al., 1993; the original idea is due to Nagao, 1984); and • …

Content truncated to 500 words