Human spatial cognition appears to operate differently in manipulable (small scale) spaces and in geographic (large scale) spaces. Although some fundamental spatial concepts may apply for both kinds of spaces, the relative salience of the concepts may be quite different. Geographic information systems (GISs) represent geographic spaces and the entities in them, but users interact with these systems as if they were manipulable, through representations that appear in manipulable spaces. This difference in scales of representation and action is not new, as people have long reasoned about geographic spaces while looking at or remembering graphical maps, which like GIS displays and equipment are manipulable entities. Part of the power and utility of maps comes from their natural space-in-space representations, but since geographic and manipulable spaces are different in how people think and reason about them, graphical maps to some extent mis-represent the geographic spaces that they show. Montello (1993) captured the essence of this dilemma when he asserted:

Even if computer representations more faithful to the nature of geographic space can be developed, users may still have to interact with these computerized databases at the manipulable scale.

This paper reviews human spatial cognition, and presents some evidence that there are at least two, and perhaps up to five or six, different kinds of spaces, distinguished by human interaction and in part by scale. Then, implications are drawn for the cognitive aspects of human-computer interaction for GIS using current technology, and speculations are made about possible interaction methods for future technologies.

2.1 What do We Mean by 'Spatial'?

'Spatial' is remarkably difficult to define without circularity. Webster's Unabridged Dictionary (1975) defines spatial as "relating to space; happening or existing in space" (p. 1740). It further defines space as "distance extending without limit in all directions; that which is thought of as a boundless continuous expanse extending in all directions or in three dimensions, within which all material things are contained" (p. 1736). Though these definitions taken together suggest that 'spatial' is about continuous, boundless, three-dimensional expanses that contain all material things, this synthesis of terms seems to 'miss the point' regarding human conceptions of space. From a geographic perspective, "place and its dimensions serve as the bases for geographic descriptions and explanations of events" (Gaile and Willmott, 1989, p. xxv). In this context, places are not boundless, nor are they readily perceived as extending in three dimensions and containing all material things. Conceptions of space, it seems, are shaped and influenced by factors that are not accounted for in the dictionary definition.

2.2 Importance of Spatial Cognition

Space is fundamental to human existence, and has a great influence on human thinking. Lakoff and Johnson's (1980) book, "Metaphors We Live By," showed how many metaphors for abstract domains of human thought are rooted in spatial experience. At an even more fundamental level, Johnson (1987) asserted that "our [image] schemata for spatial and temporal orientation are so pervasive and so constitutive of our ordinary experience that they are taken for granted (and thus overlooked) in standard accounts of meaning and understanding" (p. 31). Johnson (1987) conceived of image schemata as "structures for organizing our experience," and asserts that schemata are recurrent patterns, shapes, and regularity in "our actions, perceptions, and conceptions" thus providing meaning to "connected experiences that we can comprehend and reason about" (p. 29).

We need to act spatially at a bodily, sensorimotor scale to feed ourselves, and most people must act in geographic space to hunt, to forage for food or firewood, to shop, to commute from home to workplace, et cetera. Current human-computer interaction is certainly very spatial. The interaction from human to computer is dominated by manipulation of keyboards, buttons, mice, and perhaps touch-screens, while the direction from computers and humans is dominated by text and graphics laid out spatially on a visual display. Mark (1992) presented a typology of human-computer interaction modes that was based on general human spatial cognition and interaction, with 'direct manipulation' based on haptic (touch) and sensorimotor space, 'camera' metaphors such as pan and zoom based on pictorial spaces of vision, and transperceptual spaces, based on exploration and wayfinding experiences.

The fact that all Human-Computer Interaction is spatial may seem to make HCI for GIS and other spatial databases easier, and probably there is some truth to this supposition, but it also presents some interesting challenges, since both the spatial information about the geographic world and the 'spatialized' information in general HCI must be dealt with simultaneously through the same sorts of spatial concepts and the same sensory channels, without the user confusing one with the other.

3.1 The 'Scientific' View

The 'scientific' view of space has tended to view space as seamless and uniform. Each kind of geometry (Euclidean, Lobachevskian, etc.) is assumed to apply to all scales and to all phenomena, although it may be recognized that some are valid approximations to geometries at some scales and not others.

3.2 The 'Cognitive' View

It seems clear, however, that cognitive spaces are not organized in this way. Experiential realism, based on the works of Eleanor Rosch and others, and elaborated by George Lakoff (1987), asserts that cognitive categories and concepts typically come from human interaction with the world. This interaction is direct and straightforward only at the scale of the human body and of everyday, manipulable objects. These objects and the spaces that contain them are fundamentally three-dimensional, and relative locations are typically expressed through object-centered or viewer-centered reference frames. Recently, the term manipulable space has been used to describe these spaces, although other terms, such as 'everyday-object' space, 'table-top' space, 'haptic' space, 'sensorimotor' space, etc., have been suggested for them. Downs and Stea (1977) called them 'small-scale' or 'perceptual' spaces. Because of their three-dimensional nature, the structure of spaces or objects such as these are not easily captured in single static drawings, but often require at least two views (plan and elevation). Computer-Assisted Design (CAD) software has been developed to handle spatial information at the manipulable scale.

All researchers who recognize more than one kind of space in this sense recognize that geographic spaces are thought of differently from the spaces described above. We cannot experience these spaces holistically, but can only interact with them piece by piece, and assemble them in our minds to varying degrees through spatial reasoning and often through the use of drawn or printed maps. Downs and Stea (1977) recognized the two basic kinds of spaces. They used the terms 'large scale' or 'geographic' to refer to these larger spaces, and also called them 'transperceptual' spaces to emphasize that they are known by integrating across direct perceptual experiences. 'Geographical' is perhaps the best term for these spaces. Geographic spaces are fundamentally two-dimensional, with the vertical either thought of as an attribute of location, or ignored altogether. Relative locations are typically expressed through external reference frames based on cardinal directions or distant landmarks. Geographic Information Systems (GISs) have been developed to deal with spatial information at the geographic scale. In large part, the differences in the nature and capabilities of CAD and GIS software reflect differences in how people deal with and think about these spaces in the real world, although to our knowledge these differences have not been formally described.

Whereas it has been fairly common to distinguish these two kinds of spaces (Ittelson's (1973) object and large-scale space, Downs and Stea's (1977) small- and large-scale space, Mandler's (1983) small- and large-scale space, Pinxten et al.'s (1983) physical and sociogeographic space, Zubin's (1989) A- and D-space, Mark's (1992) haptic and transperceptual space, and Montello's (1993) figural and environmental space), some authors recognize others (see figure 3.1). For example, two models included in figure 3.1 recognize a space composed of objects larger than the human body that cannot be perceived from one perspective (Zubin's B-space and Ittelson's object space). This includes, for example, a car and a house. Though these objects cannot be manipulated and handled (therefore require locomotion to perceive all sides of the object), these objects possess the same basic cognitive properties of objects in table-top spaces.

A handful of the models shown in figure 3.1 recognize a space type whose members are spaces so large that they cannot be experienced through locomotion, but instead are learned from maps (Pinxten et al.'s cosmological space, Montello's geographical space, Muehrcke and Muehrcke's (1992) global scope, and Siegel's small-scale space). Finally, a number of models in figure 3.1 recognize spaces that can be perceived from one perspective, such as an auditorium or scenic overlook, by panning the landscape (Zubin's C-space, Mandler's medium space, Montello's vista space, Mark's pictorial space and Lynch's (1960) spatial node).

3.3 A Typology Of Spaces

Based on these models, it seems that at least five types of spaces can be identified: (1) spaces that comprise objects smaller than the human body; (2) spaces that comprise objects larger than the human body, but smaller than house-size spaces; (3) spaces perceived from multiple perspectives, piecemeal, over an extended period of time, including from inside-house spaces to city-size spaces; (4) spaces that are so large that it is impractical (if not impossible) to experience them via locomotion, including from state or country sized spaces to the Universe; (5) spaces that can be perceived without appreciable locomotion, i.e., from one perspective, including views in a room, in a stadium, at a scenic overlook. For completeness, though the following space was not considered in any of the models discussed thus far, one might also add 'microscopic' spaces too small to interact with or observe directly without special instruments.

Why have all these different kinds of spaces and typologies of spaces been introduced here? Because for most if not all human cultures, there are different kinds of spaces, applied to different situations or phenomena, that are conceptualized in different ways. It has been claimed that "a main objective of GIS is to allow the user of the system to interact vicariously with actual or possible phenomena of the world," (Mark, 1989). If users of geographic information are to be able to see through the software and hardware, then the system must present a world of spatial concepts that is as similar as possible to the concepts those people use when they reason about the real geographic world. It seems that the number of different kinds of spaces and the distinctions between them vary with culture and language. For example, some cultures use geographic (cardinal) reference frames to refer to and reason about table-top configurations (Pederson, 1993) and even to refer to parts of their own bodies (Talmy, 1983). Thus, even if the spatial concepts seem distinct within a culture, many concepts are extended by metaphor and other cognitive methods to a variety of scales, so any spatial concepts might turn up being used for any scale or for any phenomena. And lastly, as noted above, human-computer interfaces for GIS on standard computers must deal with concepts from both geographic space (the information) and manipulable space (the hardware and software).

In a task analysis context, what tasks do people perform in the real world using geographic information? The decision on where to locate a store might be made simply by driving around an area, 'looking' for good sites, relying on a near-intuitive idea of what would make a 'good' site. This might be augmented by looking at road maps on which other stores, of the same chain or competing firms, have been plotted. The paper road map can be replaced by an image on a computer screen with almost no need for conceptual change, but how can we effectively replace or model the drive through the neighborhood? Do we even want to? Virtual reality (VR) technology may be critical here, but to date most VR technology has focused on replicating manipulable and perhaps environmental spaces. Perhaps flight simulators would be a better place to look for analogs to the 'drive around' method.

The other aspect of the problem, perhaps the most interesting and challenging, is that simulation or replication of real-world interaction methods may not lead to optimal HCI methods. The map is an excellent example, since it is often easier to perform certain spatial tasks with a map than it is to perform those same tasks with access to the real world (Lloyd, 1989; Freundschuh, 1991). Catalogues of the spatial concepts involved in human conceptual models of the real world must be developed, and described formally, and for most cultures, this will have to be done at at least two different scales: the manipulable and the geographic. The spatial aspects of HCI methods in general must be subjected to similar analyses. Then, these spatial models can be compared, to search for mixed metaphors and conceptual problems that may interfere with geographic information use. Research must not stop at that level, as it is possible that entirely new conceptual models can support HCI for geographic information. Only imagination can find these.

This paper represents part of Research Initiative #13 ("User Interfaces for GIS") of the National Center for Geographic Information and Analysis (NCGIA); the NCGIA is supported by the U.S. National Science Foundation (NSF), under NCGIA grant SBR 88-10917. Thanks are especially due to Tim Nyerges, whose grants from NATO and the NSF allowed the first author to attend the workshop at which this paper was first presented.