Acquisition of Landmark Knowledge 1 Running head: ACQUISITION OF LANDMARK KNOWLEDGE Acquisition and Retention of Structural versus Object Landmark Knowledge When Navigating through a Large-Scale Space

Three studies investigated the acquisition and retention of structural and object landmark knowledge in novel virtual indoor environments. In each study, subjects were trained and tested in complex virtual indoor environments. The studies investigated the rate of acquisition and memory retention for local hallway structure (structural landmarks) or local pictures (object landmarks). These studies investigated the rate of acquisition, the role of information content and memory retention of structural (local hallway structure) and object (local pictures) landmarks when subjects were trained and tested in novel virtual indoor environments. In Experiment 1 the information content of the structural and object landmarks was the same. In this study subjects’ knowledge about the building structure was more accurate than for the object landmarks. In Experiment 2, we increased the information content of the object landmarks and found memory for the two types of landmarks to be equivalent. In Experiments 3 we tested whether the bias for remembering structural landmarks over object landmarks is due to the non-independence of the structural landmarks. The study found no evidence for this explanation. The results from these studies suggest that: a) even initially, subjects are biased toward encoding the building structure over the locations of objects; b) subjects are sensitive to the information content of landmarks and will allocate memory resources to landmarks that are more informative; and c) the memories for structural and object landmarks are resilient over long periods of time delay. Acquisition of Landmark Knowledge 3 Humans, and many other animals, possess the remarkable ability to engage in goal-directed spatial navigation behavior in large-scale spaces. That is, they can find their way from one location within a familiar environment to another, pre-determined, and unobservable location. This behavior can be accomplished without reference to external (non-cognitive) directions or representations of the space (e.g., a map) but can be achieved using an internal representation of the large-scale space. The internal representation of a large-scale space is typically referred to as a cognitive map (Tolman, 1948). Most adults carry around within them hundreds, if not thousands of these maps that allow them to easily navigate in the buildings they work in, among buildings on a campus, and in multiple cities where they have lived and visited. There is little disagreement that humans possess the ability to generate a cognitive map. What is typically debated is what is made explicit within the cognitive map and how this spatial information is acquired (Gillner & Mallot, 1998; Ruddle, Payne, & Jones, 1997; Siegel & White, 1975; Thorndyke & Hayes-Roth, 1982; Tversky, 1993). Several theories claim that landmarks play a role in the development of a cognitive map, but it is not clear which features of the environment are used as landmarks and why these features are useful when navigating. This paper will investigate this issue by defining two types of landmarks: structural landmarks and object landmarks. Structural landmarks are defined as geometric features of a layout that can serve as visual cues, such as dead-ends and T-junctions, and object landmarks are objects in the environment that are independent of its structure, such as pictures on the walls. With respect to these landmarks, the current studies will investigate three primary questions concerning the acquisition and retention of landmark knowledge. First, do people acquire knowledge about one type of landmark better than the other? Second, how does the information content of these landmarks affect encoding? And finally, is memory retention (following a delay) for these two types of landmarks equivalent? To investigate the issue of acquisition we measured subject’s landmark knowledge to see if different types of visual cues are treated differently during initial learning of an environment. Second, we manipulated the information content of these Acquisition of Landmark Knowledge 4 landmarks (i.e., how well subjects could theoretically localize themselves simply by observing a particular landmark) and measured how information content affects acquisition of the landmark knowledge. Finally, to investigate the retention of landmark knowledge we tested subjects on their landmark knowledge after delays of 1 day, 7 days, 30 days and 1 year (Experiment 1) after the initial training. The retention rate demonstrates the resiliency of the memory representations for these different types of landmarks in addition to the effect of information content on retention. Acquisition of Spatial Knowledge When developing a cognitive map of a novel environment, visual cues can be used and translated into a mental reference that can be re-used at a later time. Several theories have suggested ways in which visual information, primarily landmarks, can be used to learn an environment. One of the first spatial acquisition theories, proposed by Siegel and White (1975), states that learning the structure of an environment is based on remembering key landmarks; from this, route knowledge develops and ultimately a “map-like” survey representation of the space is generated. Having a survey representation implies having some understanding of the topography of the environment as well as the spatial relations between locations. Subsequent research suggests that route knowledge may not necessarily be an intermediary stage between landmark and survey knowledge (Moar & Carleton, 1982). It is also possible that landmarks themselves are the basis for spatial representations of environments. According to the anchor-point hypothesis, salient cues are used to “anchor” each region of a space into an organized, even hierarchical, mental map (Couclelis, Golledge, Gale, & Tober, 1987). Alternatively, the view-graph approach suggests that spatial representations consist of a series of views, or snapshots, of the environment, each associated with specific actions and goals. According to this theory, there is no need for a survey representation of space. Instead, one can simply associate a specific view with a specific action. The model moves through the environment by recognizing a view and looking up what action to generate given the view. The model then continues the action until it recognizes another view and then generates the action associated with that view. According to the authors, linking of views to actions will ultimately lead the model to its Acquisition of Landmark Knowledge 5 goal (Mallot & Gillner, 2000; Schölkopf & Mallot, 1995). Recent work by Gillner and Mallot (1998) in which they manipulated the positions of landmarks following training showed that subjects’ navigation performances were the poorest when the new positions of the landmarks offered conflicting action choices (e.g., turn left versus right). This finding suggests that movements may be associated with specific object configurations. It is also important to note that not all theories of spatial learning explicitly involve the encoding of object landmarks. Kuipers and his colleagues (Kuipers, 2000, 2001; Kuipers & Byun, 1991) have developed a robot navigation algorithm in which there is a hierarchy of representations. This algorithm, called the Spatial Semantic Hierarchy, represents space as a collection of paths that intersect at particular places. The model represents this space at three levels of abstraction: local control laws (e.g., move down the hallway), topological structure (Path-A and Path-B intersect at Place-C) and finally at a metrical level. No where in this hierarchy is object landmark knowledge encoded. Instead, the model relies on Place-Recognition that is dependent upon the local structure of the building. Recent research by Kuipers, Tecuci, and Stankiewicz (2003) has shown evidence for the topological structure proposed by this model.