behaviour might be characterised as a textual to textual transformation, where different textual representations are constructed for different purposes. These findings suggest that different mechanisms may underpin these component skills and that different forms of support for these task may be appropriate. In particular, while novices appear to rely heavily upon visualisation, experts, in contrast, adopt a more textual and closely code-based view.
Hence, one might see the evolution of expertise as a dynamic process where different mechanisms are called into play as skill develops ( Davies, 1991). One issue that has not been addressed by those who advocate different forms of support for learning programming relates to the evolution of such mechanisms. For example, possibility is that learning does not appear to be a simple linear process which demands similar, but possibly more complex forms of support for certain skills. This raises other questions that have a direct bearing upon intelligent tutoring systems work more generally. For example, when teaching certain skills, should one adopt an idealised model of expertise to which students are guided, or rather, support the manifestation of different skills as and when they appear? This may suggest a more flexible approach to tutoring where different forms of representation, whether visual or textual, are employed under different conditions.
The study presented here consisted of two conditions. In the first condition subjects were asked to generate a Pascal program corresponding to a brief specification. During the first experimental session, half of the subjects used a full screen editor (which all subjects had used extensively before) while the remainder used a restricted editor. In the second condition the first group of subjects used the restricted editor, while the second group used the full-screen editor. The subjects were asked to write a program corresponding to the specification they had received. The restricted editor was a modified version of the full screen editor, which allowed cursor movement in only one direction (from the top of the screen to the bottom) between adjacent lines.
In the second condition the same general procedure was used, however, the focus of interest in this case was related to program comprehension rather than generation. In this condition subjects were asked to study a program using either the full screen editor or a modified version of the editor which allowed the subject to read only a single line at a time. In a similar way to the generation condition, subjects were constrained to read the program in linear order. Half of the subjects used the full screen editor, while half used the restricted editor. Subjects were encouraged to take notes if they wished.
The dependent variable was the average time devoted to producing notes of different levels of abstraction. The experimenter and two other independent raters studied the notes produced by each subject. Preliminary analysis