Toward Multi-Engine Machine Translation

Toward Multi-Engine Machine Translation Sergei Nirenburg and Robert Frederlcing Center for Machine Translation Carnegie Mellon University Pittsburgh, PA 15213 ABSTRACT Current MT systems, whatever translation method they at present employ, do not reach an optimum output on free text. Our hy- pothesis for the experiment reported in this paper is that if an MT environment can use the best results from a variety of MT systems working simultaneously on the same text, the overallquality will im- prove. Using this novel approach to MT in the latest version of the Pangloss MT project, we submit an input text to a battery of machine translation systems (engines), coLlect their (possibly, incomplete) re- sults in a joint chaR-like data structure and select the overall best translation using a set of simple heuristics. This paper describes the simple mechanism we use for combining the findings of the various translation engines. 1. INTRODUCTION A number of proposals have come up in recent years for hybridiza- tion of MT. Current MT projects -- both "pure" and hybrid, both predominantly technology-oriented and research-oriented are single- engine projects, capable of one particular type of source text analysis, one particular method of finding target language correspondences for source language elements and one prescribed method of generating the target language text. It is common knowledge that MT systems, whatever translation method they at present employ, do not reach an optimum output on free text. In part, this is due to the inherent problems of a particular method - for instance, the inability of statistics-based MT to take into account long-distance dependencies or the reliance of most transfer-oriented MT systems on similarities in syntactic structures of the source and the target languages. Another crucial source of deficiencies is the size and quality of the static knowledge sources underlying the various MT systems - pmlicular grammars, lexicons and world models. Thus, in knowledge-based MT the size of the underlying world model is typically smaller than necessary for secure coverage of free text. Our hypothesis for the experiment reported in this paper is that if an MT environment can use the best results from a variety of MT systems working simultaneously on the same text, the overall quality will improve. Using this novel approach to MT in the latest version of the Pangloss MT project, we submit an input text to a battery of machine translation systems (engines), collect their (possibly, incomplete) results in a joint chart-like data structure and select the overall best translation using a set of simple heuristics. 2. INTEGRATING MULTI-ENGINE OUTPUT The MT configuration in our experiment used three MT engines: • a knowledge-based MT (K.B MT) system, the mainline Pangloss engine[l]; • an example-based MT (EBMT) system (see [2, 3]; the original idea is due to Nagao[4]); and • a lexical transfer system, fortified with morphological analysis and synthesis modules and relying on a number of databases - a machine-readable dictionary (the Collins Spanish/English), the lexicons used by the KBMT modules, a large set of user- generated bilingual …

Content truncated to 500 words