Getting Serious About Parsing Plans: A Grammatical Analysis of Plan Recognition

This paper is concerned with making precise the notion that recognizing plans is much like parsing text. To this end, it establishes a correspondence between Kautz’ plan recognition formalism and existing grammatical frameworks. This mapping helps isolate subsets of Kautz’ formalism in which plan recognition can be efficiently performed by parsing. In recent years, plan recognition has emerged as one of the best-understood frameworks for analyzing goaldirected behavior. Interest in plan recognition has led to the development of diverse recognition strategies.l One approach suggested several times is that of parsing plan descriptions (Sidner (1985), Ross & Lewis (1987)). A plan is typically described as a sequence of steps, so interpreting some observations in terms of a plan can naturally be seen as a parsing task wherein observations are lexical tokens and plan libraries are grammars. My aim in this paper is to explore this parsing view of plan recognition by establishing a formal correspondence between an existing plan formalism and contextfree grammars. By working through the details of this correspondence, the paper explores parsing algorithms for plan recognition, and delineates classes of problems for which these algorithms are applicable and tractable. Underlying this work is the plan recognition formalism of Henry Kautz (Kautz & Allen, 1986; Kautz, 1987). His approach is of particular interest because it is formal and well understood. It is also among the broadest of current formalisms, especially in the expressive richness of its plan representation. Finally, since general plan recognition in Kautz’ framework is intractable, there is intrinsic interest in identifying those aspects of his approach that cause this intractability, and those that avoid it.