From parallel grammar development towards machine translation – a project overview

[Some figures poorly reproduced in original] MT Summit VII Sept. 1999 From Parallel Grammar Development towards Machine Translation - A Project Overview - Anette Frank Xerox Research Centre Europe 6, Chemin de Maupertuis, 38240 Meylan, France Abstract We give an overview of a MT research project jointly undertaken by Xerox PARC and XRCE Grenoble. The project builds on insights and resources in large-scale de- velopment of parallel LFG grammars. The research approach towards translation fo- cuses on innovative computational tech- nologies which lead to a flexible trans- lation architecture. Efficient processing of "packed" ambiguities not only enables ambiguity preserving transfer. It is at the heart of a flexible architectural design, open for various extensions which take the right decisions at the right time. 1 Introduction Most of the existing high-performance MT systems are based on linguistic technology of the 60s, whereas research in NLP has established "higher-level" syntac- tic formalisms which allow for specification and pro- cessing of declarative, reversible grammars that assign rich structures to natural language sentences. Syntac- tic theories like Lexical-Functional Grammar (LFG), Head-Driven Phrase Structure Grammar (HPSG), or Generalized Phrase Structure Grammar (GPSG) are prime examples of declarative syntactic formalisms that assign natural language sentences, beyond a con- stituent structure, levels of representation which en- code morphosyntactic, lexical and – most importantly – functional syntactic information. This high-level functional information is valuable for high quality NLP applications like machine translation, informa- tion extraction, or other scenarios where information about "who did what to whom" is relevant. Over many years, research in Machine Translation has explored the potential of higher-level syntactic for- malisms for new generation MT technology. Linguistic research at Xerox is focusing on the theory and application of finite-state technology in NLP, as well as research in the linguistic, mathe- matical and computational foundations of Lexical- Functional Grammar. Theoretical contributions to re- search into LFG include the development of processing algorithms, principles of syntactic description for var- ious linguistic phenomena (e.g. long distance depen- dencies, coordination), and extensions of the LFG pro- jection architecture for semantic representation and translation (see Dalrymple et al 1995). Since 1995, the PARGRAM (Parallel LFG Gram- mar Development) project, a joint initiative of Xe- rox PARC, XRCE Grenoble, and the University of Stuttgart (IMS), has investigated the potential of LFG for large-scale NLP applications. PARGRAM encom- passes linguistic research in LFG-based parallel gram- mar development for different languages – English French and German, and recently Norwegian with the University of Bergen as a new partner. Along with the PARGRAM project, Xerox PARC has developed the XLE system (Xerox Linguistic En- vironment), a platform for large-scale LFG grammar development. XLE comprises interfaces to finite-state preprocessing modules for tokenization and morpho- logical analysis, as well as an efficient parser and gen- erator for LFG grammars. The XLE parser (and a new generator under development) are based on ad- vanced computational technologies. The development of the XLE system constitutes an independent re- search project into computational algorithms for pars- ing, generation, and most …

Content truncated to 500 words