Chart Generation Using Production Systems

Productions systems, traditionally mainly used for developing expert systems, can also be employed for implementing chart generators. Focusing on bottom-up chart generation, we describe how the notions of chart algorithms relate to the knowledge base and Rete network of production systems. We draw on experience gained in two research projects on natural language generation (NLG), one involving surface realization, the other involving both a content determination task (referring expression generation) and surface realization. The projects centered around the idea of ‘overgeneration’, i.e. of generating large numbers of output candidates which served as input to a ranking component. The purpose of this paper is to extend the range of implementation options available to the NLG practitioner by detailing the specific advantages and disadvantages of using production systems for NLG.