Building a Process Description Repository with Knowledge Acquisition.

Although there is an abundance of how-to guides online, systematically utilising the collective knowledge represented in such guides has been limited. This is primarily due to how-to guides (effectively, informal process descriptions) being expressed in natural language, which complicates the process of extracting actions and data. This paper describes the use of Ripple-Down Rules (RDR) over the Stanford NLP toolkit to improve the extraction of actions and data from process descriptions in text documents. Using RDR, we can incrementally and rapidly build rules to refine the performance of the underlying extraction system. Although RDR has been widely applied, it has not so far been used with NLP phrase structure representations. We show, through implementation and evaluation, how the use of action-data extraction rules and knowledge acquisition in RDR is both feasible and effective.