Pangloss: A Machine Translation Project

The project involves three sites (NMSU, USC, CMU) and is devoted to enhancing the state of the art in machine translation of natm'al language texts. Pangloss uses a hybrid, multi-engine approach, though knowledge-based machine translation takes a majority of resources. Types of work in the knowledge-based direction include: • continuing development of a set of knowledge acquisition tools and utilities to support the above work. • building high-quality syntactic, semantic and pragmatic analyzers for the source languages (Spanish and Japanese); • developing a high-quality text planner and generator for the target language (English); • designing a highly-expressive text meaning representation language (the interlingua) in which the results of analysis and input to generation are recorded; • putting togethera detailed domain model ("ontology") and, for each of the languages involved, a lexicon which maps lexical units of respective languages into elements of the world model and a grammar. Other types of work includes • developing an example-based machine translation engine , a first system of this kind to attempt processing of full text rather than specific examples; • developing a backup lexical transfer system based on a battery of glossaries, dictionaries, gazetteers, etc. and fortified with morphological analysis and synthesis routines ; • developing a mechanism for integrating the results of the multiple MT engines in a hybrid, multi-engine system; • continuing development, deployment and testing of a human-computer interface, a translator's workstation, to support postediting and interactive editing during the operation of the translation system; and 450