Research into cognitive mapping often focuses on its product, the cognitive map, rather than its processes. Golledge's (1999a,b) book is hence a timely addition to the cognitive mapping literature, focusing on what must be the primary reason for the evolution of a cognitive map, namely, the need to find one's way in one's environment. We review this book from the perspective of computational theorists in cognitive mapping. We seek explanations for the various cognitive mapping processes, i.e. we want to know what is computed, why it is computed, and how it could be computed. It is pleasing to see that some of the chapters in this book are addressing these issues.
2. Part I is meant to describe human cognitive maps and wayfinding and it has four chapters. Since Lynch's (1960) classic work, geographers have provided us with a telling tale of what is happening in our head when we move about in our own environment, mainly in cities. It was therefore pleasing to read Golledge's comprehensive review of the geographers' work in this area (see chapter 1). However, the chapter contains little in the way of new results on the nature of cognitive maps. For instance, the description of work on landmarks, routes and configurational properties of cognitive maps was based on the familiar work of geographers in the 1980's. If there have been no significant new results in recent years then it would have been reassuring if Golledge had affirmed this.
3. The second chapter is an attempt by Allen to develop a new framework to investigate why individual differences exist in wayfinding behaviour. The new framework takes into account one's general spatial abilities in addition to one's cognitive mapping abilities. The model (as discussed from page 74 to 80) is interesting but we find the preceeding seventeen pages of discussion on general spatial abilities in the various traditions to be far too long. This is especially so for a chapter which is meant to be part of the section entitled "Human cognitive maps and wayfinding".
4. Chapters 3 and 4 are concerned with the same problem, namely how do humans make their decisions when moving through an environment? However, they each offer a very different perspective. Garling, of chapter 3, focuses on how the various attributes affecting one's decision-making are utilised. His discussion on the nature of the different attributes, both spatial and non-spatial, is most useful for those who are interested in developing a model of the cognitive mapping process.
5. Stern and Portugali, of chapter 4, are more ambitious. They attempt to describe the entire process of environmental cognition and decision making in urban navigation. Unfortunately, they leave out significant details about their model. For example, on p.105, they pointed out that the model relevant to the present discussion is the intrapersonal prototype submodel and yet no description is given as to what this submodel is. On p. 110, their description of the decision field theory is too brief for us to work out what exactly is being formalized. Figure 4.6, with little explanation in the legend, does not help either.
6. Part II is about perceptual and cognitive processing of environmental information and it consists of 3 chapters. All the chapters were concerned with how well humans remember where things are when they move from one point to another, usually without vision and motion is along a straightline path with a few turns. Unfortunately the problem is studied as a memory problem involving a spatial task (albeit, one which is relevant to cognitive mapping but is not the cognitive mapping process itself). Thus, these works are interesting only if the authors extend their discussion from these experiments to the cognitive mapping task but none did (except chapter 5, which gave a brief comment; see below). As such, we find these chapters out of place in the book even though the experiments described are interesting. If the authors had attempted what Allen was trying to do in chapter 2, the chapters would have been more relevant and much more interesting.
7. Chapter 5 did attempt to draw some conclusions regarding the lessons learned from these laboratory-based experiments on path integration with respect to real-world human navigation. The latter is related to cognitive mapping. Since the experiments described suggest that humans are poor at path integration, one possible conclusion, that is noted by the authors, would be that it plays little role. However, the authors also cautioned that such a conclusion might be premature. Since chapter 5 is entitled "Human navigation by path integration", we find it puzzling to read many remarks concerning aircraft navigation and path integration by nonhuman species.
8. Section III examines wayfinding and navigation in nonhuman species and the spatial representations involved in these tasks. In chapter 8 Etienne, Maurer, Georgakopoulos and Griffin examine the role of dead reckoning and landmarks in representations of space. They postulate that for rodents, dead reckoning is a requirement for landmark learning. Dead reckoning allows the animal to keep track of its location and thus provides the reference frame for defining the location of the landmark. This stresses the importance of both an extension and location in space for the landmark, a notion which is often overlooked in many of the human behavioural studies which follow the dominant landmark to route to survey knowledge theory of Siegel and White (1975). Dead reckoning provides the animal with metric knowledge, knowledge which according to Siegel and White's theory is not available until survey knowledge has been acquired, long after the acquisition of landmarks. Alternative theories which propose the early acquisition of metric knowledge can be found in Moar and Carleton (1982), Anooshian (1996), Montello (1998) and Yeap and Jefferies (2000).
9. Judd, Dale and Collett present a view-based account of insect navigation in chapter 9. They propose that familiar scenes are remembered as two-dimensional views. The insects learn these views as visual patterns in retinotopic coordinates. A view is recognised when its elements fall on exactly the same part of the retina as when the view was initially learned. Landmarks too are stored as two-dimensional views. These are involved in two navigation strategies, beacon aiming and the linking of flight trajectories to views of beacons. But it is not clear what exactly a landmark is to an insect. Is the landmark some distinguishable part of a scene or is it the whole scene?
10. Chapter 10, by Wiltschko and Wiltschko, describes the spatial representations and navigational abilities of birds with regards to homing and migration. It is the homing behaviour of the bird which is of particular interest because this involves the construction of a cognitive map, termed the navigational map. The navigational map is a representation of gradient directions which are derived from the home region. When the bird is away from the home region it extrapolates on these gradient directions stored in its map. In addition the bird uses a "mosaic" map when it is in the vicinity of home. This is a "directionally oriented mental representation of the distribution of familiar landmarks". Wiltschko and Wiltschko point out that there are parallels in the way in which birds develop their representations and the way in which humans develop their representations as proposed in the theory of Siegel and White (1975). A significant difference according to the Wiltschkos' is that birds have compass information which means that the mosaic (landmark) map and the navigational maps are in fact survey maps. This is a fundamental difference. Siegel and Whites' developmental progression of landmark to route to survey map does not exist in the developmental progression of the avian map.
11. Thinus-Blanc and Gaunet discuss their view of cognitive maps as "active information-seeking structures" in chapter 11. From the point of view of developing a computational theory of cognitive mapping (Yeap & Jefferies, 1999), we were interested in what Thinus-Blanc and Gaunet could contribute to our understanding of this process. Thinus-Blanc and Gaunet's chapter is significant in that it sheds light on the nature of the cognitive mapping process. In particular, their clever use of the dishabituation paradigm to uncover what information attracts the attention of the hamster/rats provides us with indirect evidence as to the nature of the map being computed. Their experiments support an earlier finding by Cheng (1986) and Cheng & Gallistel (1984) where rats were found to compute a geometric representation. Our own theory of cognitive mapping stresses the need to compute a geometrical representation of local spaces early in the cognitive mapping process (Yeap & Jefferies, 1999).
12. The final section presents two very diverse topics, Nadel and Berthoz et al's neural aspects of wayfinding in chapter 12 and 13 respectively and Chown's error tolerance and generalization in cognitive maps in chapter 14. Nadel presents an overview of the neural mechanisms of spatial orientation and wayfinding outlining the contribution of the hippocampus to cognitive mapping. A review article such as this has considerable value in bringing together different lines of research in the same area. However, from the title we were also expecting to see more on how the hippocampus fits into the larger spatial system than the scant coverage it received. In Chapter 13 Berthoz et al. report on their experiments which examine the role of proprioceptive and vestibular sensory information in path integration. They also give a brief but interesting discussion on neural mechanisms which process the vestibular signals. As cognitive mapping researchers who are looking for explanations of spatial behaviour (we want to know what is computed and why it is computed) we have been frustrated that neuroscientists have so far given little attention to how representations in the brain are applied to solve spatial problems such as navigation and planning. It is good to see that Nadel recognises this problem and is hopeful that it will receive some attention in the future.
14. Chown provides good arguments in Chapter 14 for a cognitive map of "qualitative" representations which are more tolerant of the errors and uncertainty which is inherent in an agent's representation of its spatial environment. His model of a cognitive map, PLAN, is based on Siegel and White's (1975) developmental progression from landmark to route to survey map. Chown argues that the primary function of a landmark is to provide a spatial index into a cognitive map. When such a landmark is recognised it provides the clues as to where one is within an environment. Chown further argues that humans tend to conceive of space as collections of familiar landmarks. A spatial representation of landmarks is thus qualitative in the sense that it is imprecise, it does not depend on precise metric information. The notion of a landmark being identified with a region of space is appealing. Some problems, such as path planning, are much simpler to solve if one can reason with the landmarks that stand for a region of space rather than more detailed representations that could be metric. However, the idea of a spatial index appears to contradict Siegel and White's developmental progression of landmark to route to survey information. If a landmark is to represent a region of space in the way in which Chown describes, then this presupposes that initially the whole region would have been identified including "where" it is within some larger framework. Chown suggests that, in time, humans develop qualitative internal representations that have the characteristics of absolute representations of space. Perhaps some of these characteristics need to be made available before these qualitative landmarks can even be computed.
Anooshian, L.J. (1996) Diversity within spatial cognition - Strategies underlying spatial knowledge. Environment and Behaviour, 28(4): 471-493.
Cheng, K. (1986) A purely geometric module in the rat's spatial representation. Cognition, 23: 149-178.
Cheng, K. & Gallistel, C.R. (1984) Testing the geometric power of an animal's spatial representation. In H. L. Roitblat, T. G. Bever, & H.S. Terrace (eds), Animal Cognition 409-423. Hillsdale, NJ: Lawrence Erlbaum.
Golledge, R. G. (1999a). Wayfinding behavior: Cognitive mapping and other spatial processes. Baltimore: Johns Hopkins University Press.
Golledge, R. G. (1999b). Precis of "Wayfinding behavior: Cognitive mapping and other spatial processes." PSYCOLOQUY 10(036) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1999.volume.10/ psyc.99.10.036.cognitive-mapping.1.golledge http://www.cogsci.soton.ac.uk/cgi/psyc/newpsy?10.036
Montello, D.R. (1998) A new framework for understanding the acquisition of spatial knowledge in large-scale environments. In M.J. Egenhofer & R.G. Golledge (eds), Spatial and Temporal Reasoning in Geographic Information Systems 143-154 New York: Oxford University Press.
Siegel, A.W. and White, S.H. (1975) The development of spatial representations of large-scale environments, In H.W. Reese (ed) Advances in Child Development & Behavior by H.W. Reese 10, 9-55. New York: Academic Press.
Yeap, W.K. and Jefferies, M.E.(1999) Computing a representation of the local environment, Artificial Intelligence 107 265-301.
Yeap, W.K. and Jefferies, M.E. (2000) On early cognitive mapping. (in preparation).