HUMAN-COMPUTER INTERACTION
SECOND EDITION
HUMAN-COMPUTER INTERACTION
SECOND EDITION
Compared with the deep cognitive understanding required to describe problem-solving activities, the human motor system is well understood. KLM (Keystroke-Level Model [36]) uses this understanding as a basis for detailed predictions about user performance. It is aimed at unit tasks within interaction -- the execution of simple command sequences, typically taking no more than 20 seconds. Examples of this would be using a search and replace feature, or changing the font of a word. It does not extend to complex actions such as producing a diagram. The assumption is that these more complex tasks would be split into subtasks (as in GOMS) before the user attempts to map these into physical actions. The task is split into two phases:
The formalisms we have seen so far have some implicit or explicit model of how the user performs the cognitive processing involved in carrying out a task. For instance, the concept of taking a problem and solving it by divide and conquer using subgoals is central to GOMS. CCT assumes the distinction between long- and short-term memory, with production rules being stored in long-term memory and 'matched' against the contents of short-term (or working) memory to determine which 'fire'. The values for various motor and mental operators in KLM were based on the Model Human Processor (MHP) architecture of Card, Moran and Newell [37]. Another common assumption which we have not discussed in this chapter is the distinction between linguistic levels -- semantic, syntactic and lexical -- as an architectural model of the user's understanding.
In Chapter 1, we discussed some of these architectural descriptions of the user as an information-processing machine. Our emphasis in this section will be to describe a couple more architectural models that are quite distinct from those described in Chapter 1 and assumed in the earlier parts of this chapter. Here we will see that the architectural assumptions are central to the description of the cognitive modelling that these approaches offer.
Though the problem space model described briefly above is not directly implementable, it is the basis for at least one executable cognitive architecture, called Soar. We do not discuss the details of Soar's implementation; the interested reader is referred to Laird, Newell and Rosenbloom [133]. An interesting application of the Soar implementation of problem spaces has been done by Young and colleagues on programmable user models (or PUMs) [266]. Given a designer's description of an intended procedure or task that is to be carried out with an interactive system, an analysis of that procedure produces the knowledge that would be necessary and available for any user attempting the task. That knowledge is encoded in the problem space architecture of Soar, producing a 'programmed' user model (the PUM) to accomplish the goal of performing the task. By executing the PUM, the stacking and unstacking of problem spaces needed to accomplish the goal can be analyzed to measure the cognitive load of the intended procedure. More importantly, if the PUM cannot achieve the goal because it cannot find some knowledge necessary to complete the task, this indicates to the designer that there was a problem with the intended design. In this way, erroneous behaviour can be predicted before implementation.
Barnard has proposed a very different cognitive architecture, called interacting cognitive subsystems (ICS) [17, 18, 19]. ICS provides a model of perception, cognition and action, but unlike other cognitive architectures, it is not intended to produce a description of the user in terms of sequences of actions that he performs. ICS provides a more holistic view of the user as an information-processing machine. The emphasis is on determining how easy particular procedures of action sequences become as they are made more automatic within the user.
ICS attempts to incorporate two separate psychological traditions within one cognitive architecture. On the one hand is the architectural and general-purpose information-processing approach of short-term memory research. On the other hand is the computational and representational approach characteristic of psycho-linguistic research and AI problem-solving literature.
ICS is another example of a general cognitive architecture which can be applied to interactive design. One of the features of ICS is its ability to explain how a user proceduralizes action. Remember in the discussion of CCT we distinguished between novice and expert use of an interactive system. Experts can perform complicated sequences of actions as if without a thought, whereas a novice user must contemplate each and every move (if you do not believe this distinction is accurate, observe users at an automatic teller machine and see if you can tell the expert from the novice). The expert recognizes the task situation and recalls a 'canned' procedure of actions which, from experience, results in the desired goal being achieved. They do not have to think beyond the recognition of the task and consequent invocation of the correct procedure. Such proceduralized behaviour is much less prone to error. A good designer will aid the user in proceduralizing his interaction with the system and will attempt to design an interface which suggests to the user a task for which he already has a proceduralized response. It is for this reason that ICS has been suggested as a design tool which can act as an expert system to advise a designer in developing an interface.
In this chapter, we have discussed a wide selection of models of the users of interactive systems, including socio-technical and systems models and cognitive models. Socio-technical models focus on representing both the human and technical sides of the system in parallel to reach a solution which is compatible with each. SSM models the organization, of which the user is part, as a system. Participatory design sees the user as active not only in using the technology but in designing it. Cognitive models attempt to represent the users as they interact with a system, modelling aspects of their understanding, knowledge, intentions or processing. We divided cognitive models into three categories. The first described the hierarchical structuring of the user's task and goal structures. The GOMS model and CCT were examples of cognitive models in this category. The second category was concerned with linguistic and grammatical models which emphasized the user's understanding of the user--system dialog. BNF and TAG were described as examples in this category. Most of these cognitive models have focused on the execution activity of the user, neglecting their perceptive ability and how that might affect less planned and natural interaction strategies. The third category of cognitive models was based on the more solid understanding of the human motor system, applicable in situations where the user does no planning of behaviour and executes actions automatically. KLM was used to provide rough measures of user performance in terms of execution times for basic sequences of actions. Buxton's three-state model for pointing
processed in 0.004 seconds
|
| |
|
HCI Book 3rd Edition || old HCI 2e home page || search
|
|
| feedback to feedback@hcibook.com | hosted by hiraeth mixed media |
|
| |