Optimally naming commands involves maximizing the ability to convey an implicit model of system actions and relationships by choosing names which suggest those actions and relationships. Suggestiveness is hypothesized to be based upon the semantic similarity of the names and commands, which can be usefully formulated in terms of Tversky's model of featural similarity. To test this model of suggestiveness, three experiments were conducted. In the first experiment, 14 computer-naive subjects made semantic judgements about three sets of command names, and their responses were compared with judgements made by programmers about the corresponding set of editor commands. The judgements were used to create features to assign to each name and command. The suggestiveness of each name was then computed, using a simple context-free version of Tversky's similarity model. In the second experiment, another group of 12 computer-naive subjects was asked to pair the names from the first experiment with before-after pictures showing the actions of the editor commands. As expected, the frequency with which subjects picked the correct pictures was correlated with the suggestiveness of the names, with suggestiveness accounting for roughly half the variance in subjects' choices. In the third experiment, another group of 17 computer-naive subjects used an alternative method of obtaining features for the command names. Suggestiveness calculated from this second set of features produced similar correlations with accuracy. Inspection of the model's inaccuracies reveals that they are due to its lack of context sensitivity, and that simple context-sensitive versions of it will have even greater predictive power.
Journal description