Towards a Framework for Knowledge-based Machine Translation

Abstract

This paper deals with representational and computational issues coming up in the context of interfacing a German machine translation (MT) analysis and an English text generation system for translating German into English. The systems involved are the German analysis component Eurotra-D ([16]) of the EUROTRA-MT-system, a research and development project of the Commission of the European Communities (CEC), worked out at IAI of the University of the Saarland, and the English text generator Penman ([8]) developed at ISI of the University of Southern California. The research project itself and its motivations are discussed in [1], [2] and [16]. A further aspect discussed there is what MT can learn from text generation and vice versa. The former is of special interest because most of the known MT-systems are sentence-based rather than text-based. Systemic-functional linguistics (e. g. [5]) and especially systemic-functional grammar (SFG), which Penman is based upon, play an important role in the field of text-generation research (c. f. [9]). Aspects of SFG are also of importance for defining a semantic component for an MT-system as shown in the EUROTRA-D (ET-D) approach (c. f. [15]).