What does real work analysis tell us about system design?

Abstract

Human-computer interface design is explicitly or implicitly based on specific representations of the user's activity. Several strategies are used for system design. Some of them are based on "user's models" (for example: Cuff, 1980), which are not usually precise enough to predict the numerous difficulties in the system use. Others rely on general principles to elaborate command languages or to structure the man computer dialogue. These general principles may be:

• (a) the simplicity and coherence of the commands,

• (b) the "natural" aspect of the actions during dialogue,

• (c) the personalisation of the human-computer interface.

The design of command languages based on an analogy with natural languages, whether verbal (Treu, 1982; Landauer et al., 1980; Ledgard et al., 1980), graphic (Buxton, 1982) or musical (Buxton et al., 1983) does not systematically lead to solutions which will be appropriate for the functions the language is to fulfill (Fitter, 1979).

In general, these principles come from either informal observations of the use of different systems, or from the designer's intuition (Treu, 1976). This procedure may result in an incorrect representation of the operator's real activity and thus in models which have little to do with reality, leading to errors in the design. A correct representation of the activity is needed from the appropriate which characteristics of the system to be designed can be deduced. Several attempts have been made to develop general models for particular cases. Card et al., 1983, used a very detailed model of the activity of an operator while working on a word-processing task. Although this model provides precise data on the operator's behaviour, it only concerns tasks for which most of the aspects of the activity have been defined in advance. Although analysis of procedural errors is essential for the improvement of a system, this model cannot predict them.

This paper means to show that work analysis can help to elaborate activity representations wide enough to fulfill design requirements. Work analysis is not only concerned with performance, but also tries to define the structure of the activity as well as the cognitive processes of the operator within the real work situation. It aims at describing the complexity of the activity without making any a priori reduction. In order to gather relevant data from work analysis, the ergonomist must deal with a situation (man + computer + task) closely related to the one he has to design. In this case, the ergonomic work proceeds in degrees: the designer does several ergonomic experiments in the work place in order to improve the description of the functional characteristics of the system (see part one).

Work analysis can be integrated in a diffirent way: in the design process it can provide pertinent data by allowing experiments to take place in a laboratory, in order to study certain aspects of the cognitive processes involved in specific tasks. The difficulty associated with this approach depends on the choice of the variables used for the experiment. For validity's sake, the experiment must consider constraints due to both the environment and the task. But, in fact, the environmental constraints which affect the cognitive processes are often difficult to define in advance. This experimental approach proceeds by "reducing" the real situation, but this "reduction" is made "a posteriori", that is to say: after identification of the constraints due to the environment. This paper describes two examples of the ergonomic research which is being undertaken in our laboratory, each within their own context:

• (a) a close relation with a team of designers working on a particular on-line data coding system,

• (b) a more general study aiming at the design of text composition systems.

Through these two examples, we will point out the two aspects of work analysis contribution to system design.