Let me begin evaluating this objection by recalling a familiar distinction

it
206 Chapter 8
is construing spatial representation purely in terms of its behavioral
outputs.
Let me begin evaluating this objection by recalling a familiar distinction
between criterial and constitutive accounts of a particular phenomenon.
A constitutive account is intended to explain what the phenomenon in
question is, whereas a criterial account is intended to identify the conditions
whose satisfaction will be the basis (on causal explanatory grounds)
for a warranted claim that the phenomenon is indeed present. The first
point to make is that the operational and functional construal would have
to be construed as a constitutive account of spatial representation for the
objection to stick. But, of course, it is not being offered as a constitutive
account. It is a criterial account. The two different forms of explanation
should not be confused. The central tenet of functional/operational explanation
is that the best way of studying a psychological phenomenon is
in terms of the causal relations that it bears to certain inputs and certain
outputs, and it would be a caricature of this position to say that it identifies
the psychological phenomenon with these inputs and outputs. That,
after all, is the crux of the distinction between functionalism and crude
behaviorism. The sort of functional/operational analysis that I shall propose
in the next section identifies certain forms of navigational abilities
that seem to be explicable only on the assumption that the behavior in
question is governed by an integrated representation of the environment
over time. The procedure here is, of course, inference to the best explanation.
As I shall try to bring out, the empirical evidence overwhelmingly
supports the view that there is a principled class of navigational abilities
for which inference to the best explanation demands the attribution of
an integrated representation of the environment over time.
8.4 Navigation Deploying an Integrated Representation of the
Environment over Time
A good deal of the most interesting work on the representation of space
has been carried out by animal-learning theorists. There is a very good
reason for this. Animal-learning theorists study nonlinguistic creatures,
and so have no access to verbal reports as a source of evidence for the
Navigation and Spatial Reasoning 207
presence of mental representations. Of necessity, therefore, they have been
forced to consider the representational abilities of animals indirectly and
to develop very sophisticated ways of evaluating the evidence for the presence
of those cognitive abilities and testing to make sure that there are
no more-parsimonious explanations available that do not involve mental
representations. Similar points hold of those developmental psychologists
primarily concerned with prelinguistic infants. Here too we find a highly
sophisticated battery of experimental techniques designed to isolate precisely
those behavioral responses that can be explained only intentionally.
It will be no surprise, therefore, to find that animal-learning theory and
developmental psychology will feature heavily in this section.
Let me begin by stating two basic conditions that must be satisfied before
there is any possibility of ascribing to a creature integrated representations
of the environment. These are necessary but not sufficient
conditions. They correspond to the four conditions that, in chapter 4, I
placed on any behavior whose explanation requires appeal to contentbearing
psychological states. Just as these four conditions mark out those
forms of behavior that can properly be described as representationdriven,
there are two conditions that must be satisfied by any form of
navigation properly describable as driven by the representation of spatial
features of the environment (which is not the same as what I am terming
an integrated representation of the environment over time).
The first of these conditions has long been recognized to be important
by animal-learning theorists. This is the condition that genuine spatialnavigation
behavior should be place-driven rather than response-driven
(or, in terms of what is by now a familiar distinction, that it be the result
of place learning rather than response learning). The first question that
must be asked of any putatively spatial behavior is whether the apparently
spatial movement is reducible to a particular sequence of motor movements,
as opposed to being targeted at a particular place. The point of
this condition is that it maps very neatly onto the general requirement
that no genuinely representational behavior can be explicable in stimulusresponse
terms, because particular sequences of movements are paradigms
of what can be learned as responses to stimuli. The standard
stimulus-response explanation of maze learning in rats is that the animal
learns a series of responses that are reinforced by the rewards that follow
208 Chapter 8
correct responses. At each point in the maze where there is more than one
possible path (the “choice points”) the rat learns to go right or go straight
ahead, where these responses can be coded in terms of the sequences of
movements that they involve, and it is these sequences of movements that
are rewarded.
The direct reinforcement of particular sequences of movements is a paradigm
of the mechanistic explanations of behavior that stimulus-response
theory attempts to provide. This means that it will be a necessary condition
on any behavior genuinely driven by an integrated representation of
the environment over time that it not be reducible to particular sequences
of movements. But what is the difference between behavior reducible to
particular sequences of movements and behavior not so reducible? A classic
and elegant experiment by Tolman, Ritchie, and Kalish (1946) is a
paradigm illustration of the difference between place learning and response
learning. Tolman used a cross maze with four endpoints (north,
south, east, west—although these are labels rather than compass directions).
Rats were started at north and south on alternate trials. One group
of rats were rewarded by food located at the same endpoint, say east—
the point being that the same turning response would not invariably return
them to the reward. For the other group, the location of the food
reward was shifted between east and west so that, whether they started
at north or south, the same turning response would be required to obtain
the reward. This simple experiment shows very clearly the distinction between
place learning and response learning. To learn to run the maze and
obtain the reward, the first group of rats (those for which the food was
always in the same place, although their starting-points differed) must
represent the reward as being at a particular place and control their movements
accordingly. If they merely repeated the same response, they would
only succeed in reaching the food reward on half of the trials. For the
second group, though, repeating the same turning response would invariably
bring them to the reward, irrespective of the starting point.
Tolman found that the first group of rats learned to run the maze much
more quickly than the second group. From this he drew conclusions about
the nature of animal learning in general, namely that it is easier for animals
to code spatial information in terms of places rather than in terms of
particular sequences of movements. This general thesis—one that many