Community Research and Development Information Service - CORDIS

Industrial validation of a unification-based machine translation prototype

The project bridges the gap between the current academic research into natural language processing (NLP) and machine translation (MT) and the needs of industrial users. Although the user should profit from the advantages that modern NLP techniques can offer with respect to maintenance, variability and extendibility, the user interfaces and usability of the results are ultimately more important than the underlying linguistic techniques. Through intensive dialogue on both sides, the steps necessary for industrial validation have been taken without losing the advantages of a modern NLP system. The CAT2 system is an official EUROTRA sideline that runs under SICStus Prolog on a UNIX or DOS platform. The formalism is unification-based and realizes translation with a system of tree-to-tree transductors. The grammars developed within this formalism are inspired by most recent linguistic formalisms. The user produces a text in German and sends it via e-mail to IAI, where the mail triggers the CAT2 system for automatic translation into English or French. The output of the translation is sent back to the user, who via a hotline can comment on the translation and introduce, if possible, new terminology.

An extensive morphology has been developed and integrated into the CAT2 MT system for the German, English and French languages. For the three languages, three medium sized MT oriented dictionaries have been developed, where every lexical item specifies among other things the word-forming processes it may enter into (compounding and derivation) and a set of lexical functions. Language modules have been developed for the three languages based on a set of 'common' rules and language-specific parameters, allowing for the rapid inclusion of additional languages and parametrization according to the users' requirements.

Reported by

IAI
Martin-Luther-Strasse 14
66111 Saarbrücken
Germany
See on map
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top