Parallelism and the Penman Natural Language Generation System

摘要

This report discusses parallel processing for the Penman natural language generation system. We first analyze the computational requirement of the generation process. We then identify aspects of this computation that could benefit from being carried out in parallel. The Penman generator is composed of a systemic grammar, the Nigel grammar, and its environment. These two components are functionally separated and interface to each other via an inquiry mechanism. This implies that Nigel and its environment can be processed in a distributed way. We also illustrate how both Nigel and the major part of its environment, the KL-TWO knowledge base, can each be processed in parallel. In the Nigel grammar, the systems, choosers and realization statements can be activated simultaneously according to some computational dependency that resembles the system network. The KL-TWO knowledge base can be implemented as a parallel computing system and two existing approaches, using Classifier Systems and Connectionist Models respectively, are analyzed and assessed.