Abstract

"This thesis deals with the problems that arise when a human user interacts with a computer via a command language. It investigates how to make the computer behave like a helpful expert who watches over the user's shoulder and offers spontaneous help rather than simply error messages or no feedback at all, as is commonly the case. The unix command shell has been used as an example domain of this kind of human computer dialogue. A prototype active help system, called RESCUER (standing for REasoning System about Commands of Unix using Evidence Reasonably), incorporates the required additional reasoning and response strategy in the form of user modelling and advice generation components. There are three problems associated with the kind of response that is required. First the computer has to be able to judge if a situation is problematic, second it has to find the cause of the problem if there is one and third has to decide what sort of advice to give. Tracing the user's reasoning that led him/her to a possibly unwanted situation is the key to the solution of these three problems. This thesis investigates the utility of Human Plausible Reasoning theory (from now on referred to as HPR) for these purposes. HPR was originally constructed by Collins and Michalski in an attempt to formalise the reasoning people use to make plausible guesses about questions for which they do not have full information. The logic of the theory is not meant to produce rigorous proofs but to show a line of thought based on analogies, generalisations and specialisations. Here we use the the theory in two ways. First it is used from the user's point of view as a simulator of the user's reasoning to construct justifications of the commands typed to reveal the user's beliefs, both correct and incorrect. Second it is used from the advisor's point of view as a heuristic method to evaluate the typed commands and the hypotheses generated by the simulator. This thesis examines the requirements of a help system and then establishes to what extent HPR can be used for this purpose. It has been shown that if the domain knowledge is organised appropriately, HPR can in fact provide a useful framework for the generation and evaluation of hypotheses about user's possible misconceptions and can be used to reveal the deep causes of possible errors. The main focus is on theoretical issues concerning the user model, the domain knowledge representation, the modelling of a particular situation and what the role of the Human Plausible Reasoning theory can be in all these. The prototype implementation of RESCUER, acting as a human reasoning interface between unix and a user in a simulated unix environment, serves as a clear illustration of HPR's role and provides a framework which can be expanded to a useful unix aid tool."