Plan recognition in HCI: the parsing of user actions

Plan recognition could play an important role in HCI and, in particular, for advisory systems. To assess its potential utility, we conducted a “cost-benefit” analysis. The benefit is defined with regard to a major application of plan-recognition, advising for a text editor. The proportion of text-editing methods that could correctly be recognized constitutes the measure of benefit. The cost dimension is represented by the programming and computing resources required to perform the plan recognition task. The resources grow with the complexity of the formalism used. Different types of grammatical formalisms are studied. The analysis shows that almost all efficient methods can be correctly identified with a simple formalism, lexical parsing. However, inefficient methods, which are most important for advisory systems, since they indicate that advice is needed, require more complex formalisms that take semantics into account. The objects involved in methods represent in part such semantics. Objects are handled with attribute grammars. Such grammar can recognize more than 80% of the inefficient methods. They are thus the most appropriate for the application considered because they are computationally attractive and cover most of the methods.