“Almost all of human behavior can be characterized in terms of goals and plans,” write Black, Kay, and Soloway. This view informs their entire treatment of three different domains: stories, text editors, and programs.

That readers use the goal-plan approach in reading and understanding stories is supported by the following evidence: (1) Different goal-plan episodes show independence in memory recall. To give just one example, the length of an episode affects the recall of actions within that episode, but is unrelated to recall of actions in other episodes. (2) Incomplete plans for a story’s goal are filled in by the reader--who will later claim to have read the entire plan] Also, readers remember plan actions in the order needed to reach the goal, not in the order presented by the story. (3) Two stories with analogous plans and goals will cause the reader to generalize and apply the same type of plan to any similar goal.

Text editor users also organize their thoughts using plans and goals, as shown by the following evidence: (1) At the close of each subgoal in editing a text, there is a pause, presumably so that the plan for the next subgoal can be formulated. (2) As users become more familiar with the text editor, plans often found together (e.g., locate error and delete error) are chunked together, forming larger plans. The authors then go into considerable detail discussing the users’ experience level and how it affects their use of the text editor.

The third domain the paper reviews is programming. Programming differs from the other domains in that the programmer must keep track of several plans at once. Perhaps, the authors suggest, the number of active plans will turn out to be a good metric of program complexity. Evidence supporting the goal-plan approach to understanding programs includes the following: (1) Statements in sub-plans are less likely to be remembered than statements in plans higher in the hierarchy, both after insertion (recall) and before insertion (namely, a bug). (2) Programmers seeing typical constructs (plans) can fill them in. (To give one example, “total:=total+new” requires the previous statement “total:=0,” etc.) (3) Changing the order of certain statements to conform to plan paradigms improves programmer performance.

The section concludes with an interesting discussion of the transfer of planning knowledge across domains.

The paper concludes with an interesting speculation on why programming is “hell” and “torture,” while stories are fun. I leave this discussion for the reader to discover.

The paper as a whole is very interesting, carries its point successfully, is supported by well-chosen examples, and should be of interest to a large segment of the computing community.