Machine Translation for Battlefield Messages by Lexico-Structural Transfer

We propose a machine translation system for battlefield messages based on the notion of lexico-structural transfer. Lexico-structural transfer consists of associating lexical items from the source and target language, along with extensive syntactic and lexical-semantic (structural) information about the lexemes. This approach allows us to represent divergences in translations and semantic generalizations, but at the same time it allows us to exploit word-based statistical techniques for parsing and transfer. The basis of our approach is a thorough linguistic investigation of the sublanguages. encountered in this domain. We then propose to leverage existing. research on stochastic parsing by~using SuperTags or other dependency-related parsing systems that have been proposed at the University of Pennsylvania, as well as stochastic techniques for automatic text alignment and induction of bilingual lexicons. We will design and propose a component for lexico-structural transfer. Generation will be handled by our dependency-based generator shell, Cogent. We propose to assemble a functional prototype as a feasibility demonstration which will translate battlefield messages in one or two subdomains from English into a second language. Translation of sublanguages in restricted domains is a promising field for the fruitful application of emerging natural language processing technology. The applications range from battlefield messages to weather reports to financial transaction. reports and other civilian applications.