Learning where to look: the acquisition of location knowledge in display-based interaction

LEARNING WHERE TO LOOK: THE ACQUISITION OF LOCATION KNOWLEDGE IN DISPLAY-BASED INTERACTION Brian Douglas Ehret, Ph.D. George Mason University, 1999 Dissertation Director: Dr. Wayne D. Gray The locations of interface objects (e.g., buttons, menu items) are central and necessary components to direct manipulation; to be used, these objects must be located, pointed at, and clicked on. Knowledge of the locations of sought-after objects can significantly reduce the visual search space and thus reduce performance times. Research indicates that people do indeed learn the locations of interface objects and use this location knowledge to improve performance. The question of how the impetus and opportunity for location learning change as a function of the cost structure of an interface was explored via a two-phased approach; the first part empirical, and the second analytical. The empirical phase was comprised of two experiments. The first experiment employed an incidental learning paradigm in which participants perform a search and select task and were subsequently forced to rely on their location knowledge. Experiment II used the same search and select task as Experiment I, but involved the collection of eye gaze data as a longitudinal and direct behavioral measure of location learning. The results of these experiments were used to constrain the behavior of a computational cognitive model in the second phase of the project. The model, built using ACT-R/PM (Anderson, 1993; Anderson & Lebiere, 1998; Byrne & Anderson, 1998), interacts with the same experimental task as the participants. Guided by the assumption that participants acted rationally, seeking maximum gain at minimum cost, the model provided a compelling and detailed account of key attributes of participant s behavior, from fine-grained components of interaction such as eye and mouse movements to higher order measures such as performance time. An analysis of the underlying assumptions and behavior of the model yielded three primary implications for a theory of location learning: (1) locations are encoded as a by-product of attention, (2) once encoded in memory, location knowledge is subject to the same mechanisms as other declarative knowledge, such as associative learning and decay, such that, (3) the ability to retrieve location knowledge, like other knowledge (e.g., a phone number), requires repetition, practice, or explicit rehearsal. The empirical results, taken together with inferences drawn from the behavior of the model, demonstrated that location learning is not only pervasive, but also subject to the cost structure of the interface. As the cost of relying on labels to locate the currently needed interface object (search cost) increased, so did the rate of location learning and reliance on location knowledge. Likewise, as the cost of relying on an object s label to evaluate whether that object is indeed the one currently needed (evaluation cost) increased, so did the reliance on location knowledge. Consistent with a rational analysis perspective, participants came to learn and rely on location more quickly when the interface provided them with no less-effortful alternative.