Educational Technology & Society 4(1) 2001
ISSN 1436-4522

Cognitive Navigation: Toward a biological basis for instructional design

Steven Tripp
University of Aizu
CLR, Aizu-Wakamatsu 965-8580



Learning is like navigation in space.  Human declarative learning has a spatial navigation basis.  The evidence for this comes from animal navigation research and human brain studies.  The two forms of navigation correspond to two forms of declarative memory and this has implications for instructional design.

Keywords: Navigation, Instructional design, Cognition, Cognitive maps, Hippocampus


Tolman (1948) reported that as early as 1929 Lashley had observed something quite interesting about rats.  A rat that he had trained in a maze escaped near the starting point and ran across the top of the maze directly to the goal-box where the food was located.  This behavior suggested that the rat had a map of the territory, not just a trained path to the goal.  To test the map hypothesis, Tolman and his students created a series of experiments in which rats were first trained in an alley maze which forced them to go indirectly to a goal (food).  Later the same rats were put in a different maze with a sunburst pattern consisting of alleys radiating at about 15-degree intervals.  The straight-ahead alley, which corresponded to the trained path, was blocked.  The most popular alternative alley was the one that pointed almost directly to the place where the food had previously been located.  The second most popular alley was the one pointing 90 degrees to the side where the food was.  Evidently the rats had constructed some kind of representation of their environment which allowed them to take novel paths when the learned path was blocked.  Tolman called this representation a cognitive map.


Cognitive Maps and Online Learning

Online learning is spatial in two senses: first there is a virtual space of online objects: text and media.  Second, there is a cognitive domain that has surprising spatial qualities.  The spatiality of the online environment is strongly suggested by the names of two popular browsers: Navigator and Explorer.  The spatiality of the cognitive domain is less obvious, but may have a strong basis in the evolution of the brain.  If so, a deeper understanding of how animals and humans navigate may provide a biological basis for instructional design theory, especially in, but not limited to, online environments.


The Hippocampus and Navigation

Since Tolman (1948) first suggested that animals have cognitive maps, hundreds of studies of animal navigation and its physiological basis have been performed.  One result of these studies has been the conclusion that the hippocampus is a crucial part of the navigation systems of many animals.  Other studies show that the hippocampus is crucial to human declarative learning and also reveal biological learning mechanisms in the hippocampus.  The hippocampus is hidden under the temporal cortex in humans and is therefore somewhat difficult to access.  For this reason illustrations of the human brain often omit it.  However in smaller mammals and birds it is exposed and for that reason has been highly investigated.


Animal Studies

Many animals have sophisticated navigational abilities (Healy, 1998).  Some animals use landmarks for navigation. Digger wasps memorize visual properties of landmarks near their nest.  Bees and wasps tend to face in a constant direction when they are close to their nest. Ants make use of local landmarks but if those landmarks are shown prematurely they are ignored and the ants will continue home as if the landmarks were not there. However insects don't seem to have cognitive maps of their area but instead have knowledge of a narrow path.  When wasps were caught and released at places along habitual routes they return straight home, but when they were released in positions near a route but not on it, they were lost. Many insects navigate in terms of distance and direction through a process known as path integration. An animal keeps a record during its journey of the net distance and direction it has traveled from its starting point. Desert ants wander as far as 600 m from their nest, but when they find a dead insect they carry it straight home.  If the ant is displaced it will attempt to return home on the original vector ignoring local landmarks, and so miss its home. Although these small animals can perform quite complex navigational feats, it appears that they cannot form cognitive maps.

Astonishingly, spiders, bees, ants, hamsters, dogs, and humans all seem to make similar errors in their path integration (Etienne, Berlie, Georgakopolous and Maurer, 1998).  When subjects were led on an outward (upside down) L-shaped journey without obvious forms of reference they all tried to return directly to the point of departure instead of following the L-shaped path. More surprisingly, they all seemed to show a systematic bias in that they overcompensated for the outward rotation.   That is, if the outward journey follows an L to the right, the diagonal return journey shows too much right rotation.

Theories of navigation suggest that path integration may play an essential part in the construction of cognitive maps of an environment.  When rats learn a map of their environment, they rely on a spatial reference established by path integration. Thus this relatively simple navigation system is hypothesized to be the basis for more complex map learning.  Further, just as the hippocampal region is important for sun-based navigational learning in homing pigeons, the hippocampus is equally crucial for path-integration-based cognitive map learning in rats.  Therefore the hippocampus seems to mediate a transition from relatively simple spatial mechanisms such as sun compass orientation and path integration to more complex map-based navigational mechanisms. 

O'Keefe and Nadel (1978) reviewed the behavioral and neurophysiological literature and formulated the modern cognitive map theory. In this theory, allocentric coding is defined as a cognitive map that represents landmarks and places relative to reach other.  Egocentric coding is defined in relation to the animal in terms of distance and direction. . Allocentric coding explains flexible performance in many spatial tasks such as remembering locations, navigating accurately toward a hidden goal by novel routes, and taking short cuts and detours.

Evidence for the existence of cognitive maps in the hippocampus came from the finding by O'Keefe and Dostrovsky (1971) that some neurons in the hippocampuses of freely moving rats are intensely active only when the animal's head is in a particular part of its environment, regardless of the view the animal is facing.  Such neurons are called place cells.. In their book, The Hippocampus as a Cognitive Map, O’Keefe and Nadel (1978) proposed that the hippocampus was a neural structure dedicated to creating and maintaining mental maps of space.  This theory relied on two forms of evidence. First, place cells in the rat's hippocampus were selectively active when the animal was in certain locations, as we have seen above. Second, damage to the hippocampus disrupted spatial orientation.  Rats with lesions to the hippocampus could not navigate around even familiar environments.

The hippocampus is similar across different mammalian species.  "Although there is variation among mammals in the size and shape of the hippocampus, its intrinsic circuitry is very distinctive and is conserved across species.” (Healy, 1998, p.135)  Additionally, a fundamental function of the hippocampus is similar in rats and humans.  "... spatial memory in rodents, as well as conscious recollection and explicit memory expression in humans, are prime examples of fundamental declarative memory function mediated across species by the hippocampus” (Healy, 1998, p.141). However, the function of the hippocampus is not entirely spatial. Results suggest that the hippocampus is fundamentally involved in a class of memory called declarative memory.

Evidence from birds is also relevant, since, “ ...there is good evidence for anatomical homology between the avian and mammalian hippocampus” (Healy, 1998, p. 147). Birds with hippocampal lesions could not learn either landmarks or a navigational map.  The hippocampus is also involved in avian food storing.  Marsh tits store 50-100 seeds, each in a different location several meters apart, and may not retrieve this food for several days or weeks. Lesions to the hippocampuses of such birds affect their ability to retrieve food accurately while leaving intact storing behavior and the motivation to retrieve.  There are changes in hippocampal volume corresponding to seasonal changes in behavior. Hippocampal volume in the black capped chickadee varies seasonally and is at its largest during October when the birds are storing most (Healy, 1998).


Human Studies

The experimental study of the hippocampus and navigation in humans is limited by practical and ethical problems.  First, the human hippocampus is located in a region of the brain that is not easily accessed.  Second, researchers obviously cannot deliberately damage a human brain in order to determine its functions.

Non-intrusive methods have recently become available for observing brain functions in humans.  Maguire (reported in Glynn, 1999) has used PET scans to investigate London taxi drivers, famous for their navigational ability.  She asked drivers to recall routes around the city.  What she found was that the right hippocampus was particularly active during this exercise.  Simply recalling famous landmarks did not activate the hippocampus.  Astonishingly (and reminiscent of the above chickadees), further study with MRI showed that one part the hippocampus (the posterior) was enlarged in taxi drivers, and that that part seemed to grow during the time of their employment.

One way of studying the function of the hippocampus in humans is to examine cognitive deficits in those unfortunate people who have had brain damage.  The most famous such case is that of HM and his story appears in almost every account of the operation of the brain, but Glynn (1999) is recent:

In September 1953, William Scoville in Hartford, Connecticut, operated on a twenty-seven-year-old mechanic in attempt to relieve intractable epileptic seizures. …. Following the operation, the seizures were less incapacitating and HM's intelligence seemed to be slightly improved -- probably because he was less drowsy. …. The main effect of the operation was an unexpected and devastating loss of the ability to form new memories -- what neurologists call an antero-grade amnesia.  There were other problems too.  Apart from Dr. Scoville, whom he had known for many years, HM could no longer recognize any of the hospital staff, and he could not find his way about. … Formal intelligence testing showed a slight improvement, particularly in arithmetic, compared with his performance before the operation, and a whole battery of tests failed to show any deficit in perception, abstract thinking or reasoning ability.  His motivation was said to be excellent, and his personality unimpaired. ….  A few years after the operation, when HM's family moved to a new address a few blocks away on the same street, he was an unable to learn the new address -- though he remembered the old one without difficulty -- and he could not find his way home without help.  He would do the same jigsaw puzzle, or read the same magazines, repeatedly without showing any familiarity with them. ….  The researchers found that the degree of memory loss was related to the extent to which the hippocampus had been removed and they concluded that, "the anterior hippocampus and [para] hippocampal gyrus, either separately or together, are critically concerned in the retention of current experience.”….  This provides strong evidence that the hippocampus plays a crucial role in the formation of new memories; it does not, of course, imply that those memories are stored in the hippocampus, or that other parts the brain are not involved.  pp. 315-7.


The Neural Basis of Learning in the Hippocampus

It is clear that the hippocampus plays a role in spatial learning in many animals and in general memory formation in humans.  Quite a lot is known about the actual neural mechanisms of the hippocampal formation.   Bliss (1998) provides a good summary of such mechanisms.  In the 1960s a temporary change in the behavior of neurons was discovered in the hippocampus of rabbits.  Since the hippocampus is very similar in all mammals this research has implications for human learning.  Bliss found that when the main pathway into the hippocampus was stimulated, the efficiency of the pathway increased for several hours.  This increased efficiency is called long-term potentiation (LTP).  Bliss immediately noticed that such a function had the potential for information storage, i.e., memory.

LTP turned out to have other characteristics valuable for a memory device.  Two of these were input specificity and associativity.  Input specificity is demonstrated by experiments involving stimuli to two separate pathways leading to the same population of cells.  When only one pathway is stimulated that pathway alone exhibits LTP.  Associativity is demonstrated by applying a strong stimulus to one pathway and simultaneously a weaker stimulus to another pathway.  Normally the weaker stimulus would not produce LTP, but in association with the stronger stimulus the weak pathway exhibits LTP.  Bliss reports that pyramidal cells in the hippocampus can have up to 20,000 connections.  Thus, these cells can “remember,” discriminate, and associate stimuli coming from thousands of different sources.  The computational power of such an arrangement is obvious.  All the building blocks for a learning machine are present.  The chemical processes underlying LTP are well understood according to Bliss (1998), and experiments with drugs that inhibit LTP in rats at the molecular level have demonstrated results similar to the lesion studies on rats.  Navigational ability is impaired.


Cognitive Maps and the Hippocampus

Although there is strong evidence that mammals and some other animals use cognitive maps for navigation and that the hippocampus is crucial to navigation, it does not follow that cognitive maps are located in the hippocampus. Redish (1999) reviewed the vast literature on the rodent hippocampus (and other animals as well) and came to an interesting conclusion: The cognitive map can be thought of as a kind of context for reasoning about a situation.  Cognitive maps are not stored in the hippocampus, but are downloaded as needed.  When a familiar situation is encountered the appropriate context is loaded into the hippocampus.

Redish notes that two key empirical effects have driven hippocampal studies: First, place cells only show activity in a limited portion of the environment.  Second, lesions of the hippocampus in rodents degrade navigational ability and in primates (particularly humans) cause a profound antero-grade amnesia.  Each of these two effects has driven a major theory (1) that the hippocampus stores a cognitive map for navigation, and (2) that the hippocampus stores memories of events temporarily for eventual long-term storage in the cortex.

Redish notes that a rat in a maze has five possible navigational strategies:

  1. Random navigation.  The animal has no information and must search randomly. 
  2. Taxon navigation.  The animal can find a cue toward which it can move.
  3. Praxic navigation.  The animal can execute a fixed motor program.
  4. Route navigation.  The animal can learn to associate direction with each sensory view. Route navigation can be thought of as chaining sequences of taxon and praxic substrategies.
  5. Locale navigation.  The animal can learn a map on which the location of the goal is represented.  If it knows both its own location and the location of the goal in the same coordinate system, then it can plan a path from one to the other.

These five strategies can be grouped into egocentric navigation (one to four) and allocentric navigation (five).  In contrast with the first four strategies, allocentric navigation requires the construction of a cognitive map, as originally proposed by Tolman (1948) as an explanation for shortcut abilities and latent learning (where the prior experience in an environment makes later tasks easier to learn).  One of the key points that separates allocentric navigation from egocentric strategies is that allocentric navigation is an all or none phenomenon (O’Keefe and Nadel, 1978).  The animal either knows where it is on the map or doesn't.


Memory and the Hippocampus

Glynn (1999) and many others suggest human memory can be divided into declarative and procedural memory.  Procedural memory encodes motor skills and other kinds of automatic processing and is apparently not mediated by the hippocampus.  The hippocampus formation seems to be involved only in declarative memory. Declarative memory can be further divided into episodic and semantic memory. Episodic memory is memory for experienced events.  Semantic memory is memory of facts, concepts, names, etc.  Glynn (1999) suggests a correlation between the two types of declarative memory and the two types of navigation: allocentric navigation with semantic memory and egocentric (route) navigation with episodic memory:

Learning our way about a house or a town or attractive countryside is something we are rather good at-perhaps not surprisingly given the survival value of not getting lost.  ….  Initially, learning our way must depend on episodic memory, but as the area becomes familiar we do not remember the individual journeys that we have made, and a knowledge of the layout would seem to be just part of our knowledge of the world—in other words, part of our semantic memory.  p. 320.

Redish (1999) concluded after reviewing the animal navigation literature that the function of the hippocampus is reinstantiation of context upon entering a scene and localization of the animal in a coordinate frame.  In other words, cognitive maps (a form of declarative memory) are stored elsewhere but are somehow retrieved from memory and placed in the hippocampus and location information is related to the map.

Because the hippocampus represents, not just location, but location within a reference frame, it represents the general context an animal has experienced as well as the sequence of locations.  Thus when route sequences are replayed, the map within which the sequences occurred is also replayed.  This means that the hippocampus is in fact replaying the general context. p. 216-7


Human Navigation

Over the past 2000 years or so sailing peoples discovered and populated virtually all of the island groups of the Pacific.  The navigational skill involved was enormous as can be appreciated by looking at the relative isolation of the Hawaiian Islands and Easter Island.  (An account of Micronesian and Western navigation can be found in Hutchins, 1999) The prevailing winds and currents, which are predominantly from the East, make accidental discovery highly unlikely.  By the time of western discovery the Polynesian peoples had mostly abandoned long-distance voyaging and had lost the skills involved.  However, the North Pacific peoples of Micronesia continued to make long voyages and fortunately a few navigators are still alive.  Their techniques have been studied by several researchers and are now fairly well understood.  Micronesians use a variety of techniques, including reading wave patterns and the color of the sky.  Their most sophisticated techniques involve direction finding by the stars and computing distance by imaginary or unseen islands located over the horizon from their true route.  The actual techniques involved are subtle and complex, but analysis reveals that they are a sophisticated form of path (or egocentric) navigation.  Evidence for this includes that fact that they do not use maps and the fact that they cannot “reverse” their paths without considerable mental effort (which seems to them to be unnatural).

At the time of Columbus, navigation in the Atlantic was mostly coastal (i.e., path navigation).  However, the use of the magnetic compass allowed direction finding.  Latitude could be reckoned by measuring the height of the North Star at night or by measuring the height of the sun at noon.   The instruments used for these measurements were crude, but they allowed a rough estimate.  Longitude could not be reckoned and therefore position was calculated by dead reckoning (path integration).  As the centuries passed, improved astronomy and mathematics plus accurate clocks and sextants allowed reliable map (or allocentric) navigation.  Today combinations of path and map navigation are widely used.  Similar techniques were transferred to modern aviation although until recently taxon navigation by radio signals was common.  The advent of accurate GPS (Global Positioning System) equipment is now rendering almost all path navigation obsolete.


Land Navigation

Modern land navigation probably differs little from techniques used thousand of years ago, except that physical maps may be available.  Few formal techniques are used.  Lynch, in a famous book, The Image of the City, (1960) asked people how they understood the city they lived in.  Lynch found that people typically described their city in terms of five elements: paths, edges, districts, nodes, and landmarks. 

Paths are the channels along which the observer customarily moves: the streets, walkways, transit lines, canals, and railroads.  Edges are the linear elements not used as paths by the observer.  They are the boundaries between two phases: shores, railroad cuts, edges of development, and walls.  Districts are the medium to large sections of the city that the observer mentally enters.  Nodes are the strategic spots into which an observer can enter: junctions, places of a break in transportation, a crossing or convergence of paths.  Landmarks are the points of reference.  Persons more familiar with the city rely increasingly on systems of landmarks for their guides. A series of landmarks was a standard way in which these people traveled through the city.

Lynch (1960) asked people to sketch their cities. Several sequences were apparent:

  1. Quite frequently, images were developed from familiar lines of movement. ….
  2. Other maps were begun by the construction of enclosing outline, which was then filled in toward the center. 
  3. Still others began by laying down a basic repeating pattern and then adding detail. 
  4. Somewhat fewer maps started as a set of adjacent regions, which were then detailed as to connections and interiors. 
  5. A few Boston examples developed from a familiar kernel, on which everything was ultimately hung.  (pp. 86-87).

The image was not a precise rendering of the city, however.  Rather it displayed distortions due to simplification and restructuring of the parts although the sequencing of the parts was logical.  Thus it is apparent that people carry “cognitive maps” of their environment in their heads, but the maps are not photographs of the city.  They are functional representations, simplified and distorted. Also, different subjects drew different maps of the same city, reflecting different levels of experience and activities.  As can be seen from the strategies listed above, many people represent the city at an egocentric level.  Their image of the city is based around routes that are loosely connected.  Other people have integrated cognitive maps of the city, perhaps reflecting greater experience.


Information Navigation

Maglio and Matlock (1999) reported that the way people think about the World Wide Web has implications for the way that they navigate it.  Using linguistic data based on interviews, they argue that people think of the Web as a kind of physical space in which they move.  Users see the World Wide Web in terms of a cognitive map. They remember only a few of the sites they visit, but they remember landmarks and routes and they remember key nodes of information called anchor points, much like Lynch’s subjects remembered key elements of their city.  Users relied on personal routines when trying to find information. As with maps of physical space, personal routines correspond to the routes that individuals use to get from one anchor point to another.  Analysis of language data indicated that inexperienced users talk about the Web as if it were physical place.  The data also revealed a striking distinction between experienced and beginning users.  Beginners more often referred to their bodily experiences using the keyboard, mouse, and other elements of the actual situation, whereas experienced users did not.  In addition, beginners were more likely to refer to the Web as a container than were experienced Web users. In other words, beginners may be more conscious of their physical environment, but experienced users are less focused on the specifics of the situation, as they can navigate by a more general cognitive map.  Maglio and Matlock endorse Dieberger’s (1997) city metaphor for information navigation that blends ordinary spatial elements with imaginary elements that provide semantic links as well as spatial links.  Maglio and Matlock use Fauconnier's (1997) theory of conceptual blending in information space to explain this understanding.  Fauconnier’s theory specifically links cognitive linguistics with cognitive space by positing mental spaces that are used to make sense of utterances.


Evaluating Navigational Space

According to McCall and Benyon (1999), the usability of a system depends largely on the ability of the users to engage in three activities within the information space: (1) Exploration; where there is no explicit destination. (2) Wayfinding; where a specific destination is required. (3) Object identification; which is concerned with the understanding the structure of the information space.

Wayfinding involves travel towards a specific goal and is typically a four-part activity: (1) Orienting oneself; the person seeks to locate where they and their desired destination are in the environment.  (2) Choosing the correct route; where the route decision is based on the person being correctly oriented, however the ultimate choice of route is based on various considerations including distance, direction and social and personal preferences.  (3) Monitoring the route;  once the person has embarked on a route is essential that they can tell how far is left for them to travel.  (4) Recognizing that the destination has been reached. 

Spence (cited in McCall and Benyon, 1999) suggests that wayfinding thus leads to the four following concepts: (1) Browsing; similar to exploration but the person has an imprecise goal.  (2) Context modeling; the creation of a cognitive map of the environment.  (3) Gradient perception; the weighing up of various route options. (4) Strategy formulation; the formulation of a plan used by the person when browsing.

McCall and Benyon (1999) assert that as people navigate within an environment they rely on three forms of knowledge; landmark, route, and survey (i.e., map) knowledge. Landmark and route knowledge are both egocentric.  Survey knowledge is allocentric.  McCall and Benyon identify at least three types of navigational problems. (1)  Navigational disorientation; a person can be seen looping or taking inefficient paths.  (2)  Embedded digression problem; the screen may become disorganized and cluttered.  (3)  Art Museum problem; the student does not study any part of the environment for a period of time.  Instead they merely scan it.  McCall and Benyon have developed ENISpace, navigational evaluation software consisting of a checklist of characteristics a software environment should incorporate in order to support navigation.  This checklist could be used to evaluate instructional software.


Instructional Strategies

Instructional design should be compatible with natural cognitive processes.  As a first approximation, a theory of instructional design should treat a content domain as a space for cognitive navigation.  Just as real spaces are learned first through egocentric navigation and represented in episodic declarative memory and then later represented as cognitive maps in semantic declarative memory, instructional design should allow a progression from path navigation to map navigation.


Tree maps of curriculum structure

There are many ways to represent information spatially (Herman, Melançon, and Marshall, 2000).  Many of those representations involve 2-D and 3-D graphs, such as trees (balloon trees, H-trees, radial trees, cone trees, etc.).  All of these, although attractive at first glance, suffer from flaws such as excessive complexity and data overlap.  Possible solutions such as hyperbolic trees and fisheye views introduce new, computational problems.  Fortunately, there is a relatively simple and easy-to-read form of visualization called a treemap (Shneiderman, 1992) that is adaptable to instructional purposes.  The treemap represents levels of a hierarchy with bounded rectangles.  The size of the rectangles corresponds to some salient aspect of the database.  The color of the rectangles at the terminal level represents some level of a variable.  Treemaps can represent up to seven levels of a hierarchy and thousands of nodes without a loss of usefulness.  Actual user studies (Shneiderman, 2000) indicated that smaller numbers are preferred however.  Other studies have shown that users can interpret and use treemaps quickly and find them useful.

We can represent a body of instructional knowledge (a course) with a treemap.  A course is a structured set of data.  There is a fortunate metaphorical aspect to treemaps.  They resemble city maps.  Just as Lynch (1960) found that cities have districts, nodes, edges (boundaries), and paths, tree-maps of a course have defined areas (districts and edges), nodes and landmarks (terminal documents) and (point and click) paths. It is not clear which numerical aspects of courses should be represented spatially, but the following are all possible dimensions: (1) Teacher-perceived difficulty, (2) Student-perceived difficulty, (3) Time to completion, (4) Prerequisite order, and (5) Teacher-assigned weight.


Techniques for Organizing Information Space

Waterworth (1999) suggested an information islands model in which the world (through which the structure of a set of information is presented) is seen as a group of archipelagos, each composed of information islands.  Each archipelago represents a set of related entities, providing a top-level classification of what is available in this world and where it is to be found. Each island consists of only one class of information but may contain one or more buildings and each building may contains a set of information sources related to a particular topic or application focus. Navigating between buildings and the world may be inconvenient.  In such a case, the disadvantages of the spatial metaphor can outweigh the advantages. To overcome these problems, the concept of private vehicles was developed; these can be thought of as transparent, mobile, personal workspaces.

Waterworth is engaged in the experimental design of an environment called SchemaSpace (cited in Waterworth, 1999).  SchemaSpace is a three-dimensional virtual environment in which a user may organize a collection of references to information sources located on the Web or elsewhere. SchemaSpace has been designed so as to allow the user to have four different kinds of experiences that are about different qualitative aspects of the information space:

  1. Distinctiveness: Which of the information references belong together?
  2. Quantity: How does the number of references compare to other collections found in the information space?
  3. Relevance: Of what relevance is each individual reference in relation to the subject or category?
  4. Connectedness: How do different collections of references relate to each other?


Cognitive Complexity and Cognitive Navigation

Spiro, Feltovich, Jacobson and Coulson (no date) have developed an instructional design theory called cognitive flexibility theory that is broadly compatible with the idea of navigation in cognitive space.  Cognitive flexibility theory makes recommendations about approaches that range from organizational schemes for presenting subject matter to multiple representations of knowledge (e.g., multiple classification schemes for knowledge representation).  The main metaphor they employ in the instructional design model is that of the crisscrossed landscape. They suggest a nonlinear and multidimensional traversal of complex subject matter, followed by a return to the same place in the conceptual landscape on different occasions from different directions, thereby teaching students the importance of considering complex knowledge from many different perspectives. Spiro et al. fail to distinguish between egocentric learning and allocentric learning and do not specify a relationship between the two, but it is clear that their goal is allocentric learning and they have suggested specific strategies to facilitate that goal.


MetaMaps for evaluation

Nicholson and Johnson (1999) have suggested a spatial technique for assessing understanding of complex knowledge domains.  A MetaMap is a system for recording and editing a navigational path through a corpus of data.  In some ways, this resembles Lynch’s city maps.  Different students at different levels of knowledge development will display different maps of the cognitive territory they are exploring.  An advantage of the MetaMap process is that learning and assessment are seamlessly linked.  The process of learning (note taking in MetaMap) creates an object that is then evaluated.

Nicholson and Johnson hoped to discriminate between “surface-approach” and “deep-approach” learners.  Actual results were mixed.  About half of the maps were essentially linear, perhaps indicating egocentric or path navigation and therefore a low level of knowledge integration.  Others were multidimensional with about one-third showing reflection of the issues involved, perhaps indicating the development of cognitive maps.  Student reactions were favorable although some complained that linking and annotating distracted them from developing the content.  Nicholson and Johnson were unable to confirm that the complaining students were the ones showing a “surface” learning strategy.



I have tried to demonstrate that there is a part of the brain, the hippocampus, that underlies many important forms of animal and human cognition.  From an experimental point of view it is clear that damage to the hippocampus can interfere with navigational ability in many animals.  It is also clear that damage to the human hippocampus can interfere with declarative learning.  Although the hippocampus has the neural machinery for complex learning, it is not the ultimate storage location.  It appears that cognitive maps, downloaded from elsewhere, act as a kind of context into which an animal’s location can be inserted.  How cognitive maps are constructed is not known, but it is likely that the process involves the integration of multiple episodic memories that are the result of egocentric (path) navigation.

Instructional designers face a dilemma.  The goal of instruction is often an overall understanding of a content domain.  However, the most vivid and memorable experiences are episodic.  Additionally, from an instructional point of view, path navigation is both easy to teach and easy to assess.  It is not surprising that many courses begin and end with path navigation.  Theorists warn about the shallow understandings that result from such approaches, but often they suggest that path navigation should be scorned and replaced with map navigation; i.e., rote learning should be replaced with deep understanding.  If the above account of the biological basis of animal navigation is valid, then we can extrapolate to instructional design.

Instead of emphasizing only path or only map navigation, instructional designers should present students with a domain for exploration.  The domain must have paths, nodes, boundaries, and landmarks.  Students should be encouraged to explore its various districts within which vivid activities and information may be embedded.  But simply exploring a domain is not enough.  Students must be encouraged to construct cognitive maps of the domain.  Assessment must look beyond egocentric navigation (which is wholly necessary and important) to allocentric navigation that allows flexible performance in many tasks such as remembering locations, taking short cuts and detours, and navigating accurately toward a hidden goal by a novel route.



  • Bliss, T. (1998). The physiological basis of memory. In S. Rose (Ed.) From brains to consciousness? London: Penguin Books, 73-93.
  • Dieberger, A. (1997). A city metaphor to support navigation in complex information spaces. In S.C. Hirtle & A.U. Frank (Eds.) Spatial Information Theory:  A theoretical basis for GIS (COSIT’97), Berlin: Springer Verlag, 53-68.
  • Etienne, A. S., Berlie, J. Georgakopoulos, J. & Maurer, R. (1999). Role of dead reckoning in navigation. In S. Healy (Ed.) Spatial Representation in Animals, Oxford: Oxford University Press, 54-68.
  • Fauconnier, G. (1997). Mappings in thought and language, Cambridge: Cambridge University Press.
  • Glynn, I. (1999). An anatomy of thought: the origin and machinery of the mind, London: Weidenfeld & Nicolson.
  • Healy, S. (1998). Spatial Representation in Animals, Oxford: Oxford University Press.
  • Herman, I., Melançon, G. & Marshall, M. S.  (2000). Graph information and navigation in information visualization: A survey. IEEE Transactions on Visualization and Computer Graphics, 6, 24-43.
  • Hutchins, E. (1999). Cognition in the wild, Cambridge: MIT.
  • Lynch, K. (1960). The image of the city, Cambridge: MIT.
  • Maglio, P. P.  and Matlock, T. (1999). The conceptual structure of information space. In A. J. Munro, K. Höök & D. Benyon (Eds.) Social Navigation of Information Space, London: Springer, 155-173.
  • McCall R. & Benyon, D. (1999). ENISpace:  Evaluating navigation in information space.  In  P. de Bra & J. Leggett (Eds.) Proceedings of WEBNET’99, VA: AACE, 1344-1345.
  • Nicholson, P. & Johnson, R. (1999). MetaMaps: Assessing understanding of large complex or distributed knowledge domains. Education and Information Technologies, 4 (3), 297-312.
  • O’Keefe, J. & Dostrovsky, J. (1971).  The hippocampus as a cognitive map:  Preliminary evidence from unit activity in the freely moving rat.  Brain Research, 34, 171-175.
  • O’Keefe, J. & Nadel, L. (1978).  The Hippocampus as a cognitive map, London: Oxford.
  • Redish, A. D. (1999). Beyond the cognitive map: from place cells to episodic memory, Cambridge: MIT.
  • Shneiderman, B. (1992). Tree visualization with three-maps: A 2-dimensional space filling approach. ACM Transactions on Graphics, 11 (1), 92-99.
  • Shneiderman, B. (2000). Treemaps for Space Constrained Visualization of Hierarchies,
  • Spiro, R. J., Feltovich, P. J., Jacobson, M. & Coulson, R. L. (no date). Cognitive flexibility, constructivism, and hypertext: random access instruction for advance knowledge acquisition in ill-structured domains,
  • Tolman, E. C. (1948). Cognitive maps in rats and men. The Psychological Review, 55 (4), 189-208.
  • Waterworth, J. A. (1999). Spaces, places, landscapes and views: experiential design of shared information spaces. In A. J. Munro, K. Höök & D. Benyon (Eds.) Social Navigation of Information Space, London: Springer, 132-154.