Detecting and modelling users´ beliefs about Unix

Users of the Unix operating system have misconceptions about Unix commands, make errors in specifying commands, and have incomplete knowledge of the domain. The authors´ approach is to infer users´ intentions from the commands that they issue, and use this information to detect their problems. An active chart parser generates all possible parses according to a grammar of typical plans that users follow. The points when the user might need advice are detected from a heuristic analysis of the chart, and the goal that he is attempting to achieve is inferred. This goal is verified against a STRIPS-like model of the user´s beliefs about Unix commands, which determines the misconceptions and errors that the user could possess. This model of the user´s `command beliefs´ is specific to an individual, and is dynamically altered as the session progresses to account for the user commands