LOGON: Towards a Machine Translation System Integrating LFG and HPSG

Victoria Rosén, Helge Dyvik, Dan Flickinger, Dorothee Beermann, John Carroll, Lars Hellan, Janne Bondi Johannessen, Jan Tore Lønning, Paul Meurer, Torbjørn Nordgård, Stephan Oepen and Erik Velldal

Abstract

Proceedings of LFG05; CSLI Publications On-line

LOGON is a research project in the Norwegian Research Council's KUNSTI program (Knowledge Development for Norwegian Language Technology). While including such goals as language resource development and competence building, the central task of LOGON is the development of a machine translation demonstrator translating tourism-related texts from Norwegian into English. The system translates not only between two languages but also between two theoretical frameworks: the analysis of Norwegian is based on the Norwegian LFG grammar NorGram developed on the XLE platform in the ParGram project (Butt et al. 2002), while the generation of English is based on the English HPSG grammar ERG developed on the LKB platform (Flickinger 2002). Transfer is defined over representations in Minimal Recursion Semantics (MRS), a flat semantic representation consisting of a bag of Elementary Predications and allowing underspecification of scope relations (Copestake et al. 2003). Both grammars derive MRS representations for sentences; thus MRS becomes the lingua franca for communication between the LFG and HPSG grammars.