In general, mental models seem to be cognitive views that humans construct in order to try to make sense of how things work. These views tend to be translated initally into highly visual (more generally, multi-sensory) representations in order to facilitate their communication. But the term mental model has also been connected to logical formalisms used in mathematics, computer science, and other formal disciplines in order to describe a wide variety of semantic concepts.
Beat Liver pointed out that the distinction between the presentation of information to build mental models of software and the mental models themselves is typically not clearly made. Specifically, we might call a white board diagram a mental model, when in fact it is simply an attempt at representing the mental model we have in our heads at the moment. This was in fact continually done throughout our discussions. We can only facilitate the construction of such models. For this task, we need a representation suitable for the computer (to manipulate the representation) as well as representations (at the level of the user interface) presented to a class of user (they might be different) [AL95]. For instance, Steve Edwards work on ACTI, a formal model of software subsystems provides the foundations for the former, whereas the spiral model of a heapsort implementation, discussed subsequently in this report, is an example of the latter.
Larry Latour said it might be worth while to look at interfaces from two perspectives. A Micro-architectural view is from the perspective of the knowledge inherent in that interface. A mental model view is the way different humans interpret that interface, their view of the interface semantics, colored heavily by ability, experience, and the context of a problem to be solved. This seems to be a more complex, much less formally defined way of looking at semantics, but an extremely important one nonetheless.
An example of the latter point of view was given by the Photoshop interface, crafted to be used within the context of the specific domain of digital imaging. This example and others like it reminded Robert Biddle of task analysis in the Human/Computer Interface arena. For example, such analysis is used to study the mental models that secretaries have of office software such as word processing, spreadsheet, accounting, and related business office related software. Such mental models might differ in scope and complexity from other mental models of those same software packages. A scientific modeler uses a spreadsheet interface in a somewhat different way than an accountant (although both most likely will be performing some sort of sensitivity analysis across multiple data points).
The concept of mental/conceptual models has already been discussed to a great extent in the cognitive psychology and HCI disciplines. Specifically, Don Norman has discussed these ideas in The Design of Everyday Things . The difference here is that we are interested in ``operationalizing'' these ideas to aid us in the development of software architectures by humans.