An Advanced Traveler Information System (ATIS), a key component of Intelligent Vehicle highway Systems (IVHS) in the near future, will help travelers find locations of restaurants, lodging, gas stations, and rest stops. On typical ATIS displays, which are now being incorporated in some advanced vehicles, the choices for these traveler services are presented to the vehicle occupants alphabetically. An experiment was conducted to determine whether individualizing the display through the use of neural networks enhanced performance when choosing restaurants. The neural network ATIS was compared to an ATIS that displayed the most frequently chosen restaurants at the top, one that alphabetized the list of restaurants, and one that randomly displayed the restaurant choices. The time to choose a restaurant was significantly faster for the individualized displays (neural network and frequency) when compared to the nonindividualized displays (alphabetical and random). When the two individualized displays were compared, choice time was significantly faster for the neural network approach.

The existence of design trade-offs between usability features for consumer products has been discussed in textbooks and other publications for many years (e.g., Norman, 1986). Tradeoffs occur because optimizing one feature will mean that another feature will suffer. A single product cannot be best across all dimensions of usability. In this study, NGOMSL (Natural Goals Operators Methods and Selection Rules Language) was used to model the usability of several consumer products (telephone answering machines, VCRs, calculators, and electronic organizers). In each product category, at least four different products from different manufacturers were modeled. The results showed that some product categories exhibited distinctive tradeoffs among usability features and other product categories did not. Reasons for these differences are discussed.

When performing a computer task, a user will decompose the task into cognitive goals and subgoals. These goals are accomplished through the use of external operators (e.g. keystrokes, mouse button presses) or internal mental operators (e.g. reading parts of the display, deciding on the goal). Users may utilize different goals and sequence the goals differently to accomplish the same overall task. Determining the goals and the sequencing of the goals could be useful for several reasons, such as providing a means for on-line assistance with the task. Determining these goals in the past, however, has been a time-consuming process. A neural network tool for automatically identifying cognitive text-editing goals from operators is investigated. The first of three memos edited by subjects was used to train the neural network successfully to map the operators (keystrokes) to cognitive goals. In a test of the trained network's ability to generalize to new input?the second and third memos edited by the subjects?the net could identify the cognitive goals with an overall performance accuracy of 96%. Two methods were used to investigate the validity of the goals which were identified by the tool. The characteristics of the goals were consistent with that which could be expected based upon previous research. This research illustrates that a neural network tool can identify the cognitive goals of a task.

One of the main methods used to compare direct manipulation and command-based interfaces is to examine user preferences. User preferences are extended in our research to examine which mental model, direct manipulation or command-based, subjects prefer to use in transfer situations. One group of subjects was provided with both a command-line and direct manipulation interface during a training phase. After training, several transfer experiments were conducted to determine the preferred mental model. The developed mental models were investigated by "running" the models and extending the models to new situations. They were evaluated by determining the operators specified by the subjects. By comparing the class of operators to subjects trained on only one of the interfaces, direct manipulation or command-based, the preferred model could be determined. The preferred mental model for "running" the model was the direct manipulation. For extensions that had a concrete or graphical basis in the interface, the direct manipulation was preferred. For extensions that were abstract, one model was not preferred over the other. Some ability to use multiple mental models, based upon the task, was also observed.

Consistency in human-computer interaction tasks is usually considered in a transfer paradigm in which the higher the similarity between two tasks, the higher the transfer and consistency. In a different paradigm, when tasks are performed in alternating sequences, similarity of tasks means that the mapping of interface methods or rules to overall task goals will be varied. A neural network simulation demonstrated that the same two tasks could be considered consistent in a transfer paradigm but have low consistency when they are alternated. A model called text-editing method (TEM) was developed to quantify consistency between two tasks based upon replacement, insertion, or deletion of methods. Two experiments tested the predictions of the quantitative analyses of consistency for alternating tasks. The results confirmed that similarities of two tasks and, thus, variability in the mapping of methods to overall task goals could have a detrimental effect on performance when the two tasks were performed in an alternating sequence.

The declarative and procedural knowledge, or cognitive strategies, used by expert and novice CAD operators on a 3D design task are determined in order to understand how training and management of CAD operators could affect this knowledge. Results indicate that novices were variable in performance not because of differences in declarative knowledge (on which they were trained) but because of differences in procedural knowledge (on which no training was given). The design expert could transfer procedural knowledge from other systems to the CAD system tested. The system expert could perform fast, because of the highly developed declarative knowledge, without thinking about the strategies. This research indicates the need for attention by managers of CAD systems to procedural knowledge training for system experts and novice CAD operators; declarative knowledge training can be emphasized for design experts. >

An important consideration in the design of any user interface is the relationship between the displayed information and the mental model that results after interaction with the interface. This relationship was investigated in a previous set of eight experiments for a second-order tracking task. Mental models for the task were determined through transfer experiments that forced the operators to apply their mental models to novel situations. The results indicate that, when interacting with displays containing different augmenting cues, the mental models developed are different from each other. The possibility is investigated of representing individual subjects' mental models of this system as series of rules. An experiment is reported in which the rules are generated from the collected data. Results indicate that the number of rules exhibited and the quality of the rules are dependent on the kind of augmenting cues used during training. More rules were exhibited for those subjects trained on a consistent augmenting cue than those trained on an inconsistent augmenting cue or for those trained in a no-cue control condition

The effects of internal models and tracking strategies on workload were investigated in a dual-task, second-order tracking and auditory detection experiment. Internal models and tracking strategies were manipulated by providing subjects with augmenting cues. A control group was compared with two groups provided with different kinds of display augmentation, parabola or point cues, during single-task tracking training. The display augmentation had the effect of changing tracking strategy as subjects practiced on cued and noncued trials, and it had an effect on the internal models developed. Both point and parabola augmentation reduced workload when displayed on the cued trials. On the noncued trials, the parabola augmentation, training had the effect of reducing the workload, compared with the point augmentation, even if the parabola cues were not displayed. A control group that did not change strategy during training also had low workload requirements in the dual task. The results indicate that a consistent tracking strategy or the development of a visually based internal model is needed to lessen the workload.

A current trend in cockpit design is to incorporate synthesized speech to present secondary information. Multiple-resource theories of information processing support this, but theories of stimulus/central-processing/response compatibility suggest that spatial information presented visually may have some advantages over speech if the response is manual. Two experiments compare response performance over single and dual tasks when information was presented pictorially and by speech. Pictorial subjects responded more quickly than did speech subjects. The addition of the visual tracking task in the dual-task condition had a differential effect on performance, depending on the modality of the primary task and the rate at which information was presented. The dual task impeded performance more in the fast and medium presentation rates for the speech condition but had little differential effect across rates for the pictorial condition. Analysis of the error data indicated that subjects in the pictorial condition were better able to maintain the context of the emergency than those in the speech condition. Results are discussed in terms of current theories of information processing.

The review is based on an analysis of current literature of expert systems and of system engineering models in dynamic process control. It starts with an analysis of the mental operations and cognitive requirements needed for supervisory control. Mental models are discussed as a function of situational requirements as well as of personal strategies. Systems engineering models and expert systems are briefly described and their function as decision support tools evaluated. Criteria are the overall functionality, similarity of knowledge bases and reasoning strategies of the human and the support system, adaptability to the operator's skill level and self-explanation of the support system in the interaction mode. As a result, system engineering models are only of limited value for knowledge-based process control. Expert systems seem to be very valuable tools for augmenting human decision making in process control, if the interaction problem can be solved.

A current bottleneck in the automation of cognitive tasks, such as software development, is the lack of available, standardized, reliable and valid methods for extracting knowledge from experts. This paper discusses the development of Computer Aided Protocol (CAP) to automatically collect the general and specific cognitive task components of subjects performing a programming task. The effectiveness of CAP is evaluated in a statistically balanced experimental design (n = 30) by comparing it to traditional protocol analysis and a control group. Results indicate that while neither treatment significantly altered the solution process, CAP was able to collect the lower level commands while protocol analysis collected only 56% of these lower level commands. However, protocol analysis was able to obtain significantly more high level goals than CAP. This work suggests that the integration of both protocol and CAP for knowledge extraction would provide more effective information for the development of expert systems than is feasible with either system alone.

Comparaison avec les experts humains des methodes pouvant etre utilisees pour automatiser les tâches cognitives des operateurs de commande