ABSTRACT
The work of various research groups over the last few years (e.g. Brown et al. 1990; Catizone et al. 1989; Gale & Church 1993; Kay & Röscheisen 1993) illustrates a desire to demonstrate the effectiveness of processing natural language without the need for human intervention, particularly with respect to Machine Translation (MT). This is a very attractive idea as it means that much time can be saved in creating lexical and grammatical information before the system can begin translating. In contrast, purely sub-symbolic processing simply entails presenting a system with sufficient bilingual material from which the system can induce the information necessary to translate source language input.
