ABSTRACT
In the past, most systems in knowledge-based natural language processing relied on reverse knowledge engineering (e.g. Cullingford, 1978; Dyer, 1983; Lehnert, 1988; Wilensky, 1978). By carefully examining a domain and a task, knowledge representations and control structures were developed to produce the desired behaviour. In the last decade it has become more obvious that the development of appropriate knowledge structures, their communication, and their control, needs an enormous effort in knowledge engineering (see for example Dyer, 1983). Furthermore, the complexity of the representations and their control makes it difficult to scale up the memory and the language processing capacity of these systems (DeJong, 1979; Dyer, 1983; Lehnert, 1988).
