A second point is that there seems room for an unanalyzable feeling of

reading.
182 Chapter 7
A second point is that there seems room for an unanalyzable feeling of
familiarity as a basic building block for cognition. The notion of familiarity
is at the heart of the theory of memory that Russell offered in The
Analysis of Mind, and it has by and large not received good press among
philosophers (Russell 1921, Pears 1975). According to Russell, the feeling
of familiarity integral to the content of memory images gives rise to the
belief that the image refers to the past. One significant, perhaps conclusive,
difficulty for this theory is describing the feeling of familiarity in a
way that clearly distinguishes it from the belief to which it is supposed to
give rise, and the result has been a general distrust of appealing to feelings
of familiarity in the context of memory. But this general distrust must be
mistaken. An enormous part of learning, and hence of cognition as a
whole, depends upon grasping that one event is similar to another, that
something is the same as something else, and at the most basic level, it is
hard to see how this could ever be apprehended without some sort of
feeling of familiarity. Learning depends on the appreciation of similarity,
as Quine has noted on many occasions, and familiarity is just an appreciated
similarity. This is certainly not compromised by granting the anti-
Russellian point that, in the case of belief-forming and language-using
creatures, this feeling of familiarity is not clearly distinguishable from beliefs
about the pastness of the remembered event to which it gives rise,
since we are interested in familiarity as experienced by creatures incapable
of entertaining beliefs about the pastness of past events.
Third, there is a close relation among recognitional abilities, features,
and basic inductive generalizations. When an animal recognizes a particular
place through recognizing a particular feature, that feature will, of
course, be something in which it has a direct interest and that is directly
relevant to its behavior. Recognizing a given feature is closely bound up
with recognizing possibilities for action relative to that feature, and when
this is performed over a period of time, it is appropriate to speak of the
formation of a basic inductive generalization about what to expect at a
particular place. The induction is generated by the experience of repeated
conjunctions (which is itself presumably underwritten by feelings of familiarity).
This gives us a further important element in nonconceptual
recognition, because the expectancies to which such inductive generalizations
give rise are presumably part of what it is like to recognize a place
in terms of a feature.
Points of View 183
These basic inductive generalizations are to be distinguished from
the instrumental protobeliefs discussed in chapter 5. Those instrumental
protobeliefs, involved in explaining the simplest and most basic forms of
intentional activity, were elucidated in terms of present-tense expectancies
about the behavior of objects in the environment. They do not count as
instances of conscious memory. In fact, they fail to satisfy the criteria for
conscious memory in the second of the two ways identified earlier. Although
instrumental protobeliefs have, of course, been produced by past
experiences, they contain no reference back to the past experiences causally
responsible for those effects.
There is room for a strong claim about the connection between recognitional
abilities and basic inductive generalizations at the nonconceptual
level. There is a very important question about how memories can have
have implications for current behavior. Jonathan Bennett has posed the
question and solved it (note, though, that his terminology is rather different
from mine):
Non-linguistic behaviour is essentially a manipulation of a present environment,
and the immediately relevant beliefs must be ones about the present. If an item of
behaviour of this kind is also to show that the agent has a certain belief about the
past, this must presumably be because the attribution of the past belief helps to
explain the present one. But that requires that the two be somehow linked by the
agent, and I do not see how they can be linked except through a general belief.
(Bennett 1976, 104)
It is through the medium of basic inductive generalizations that a creature
can bring its conscious memories to bear upon current behavior. Here is
Bennett’s illustration:
Suppose that a chases a victim v up a tree, guards the foot of the tree for a while,
and then climbs up after v. We could discover that a climbs the tree because he
believes that v is in the tree. Can that belief be explained as arising from the belief
that v was earlier in the tree? Only, I suggest, if a is also credited with the linking
belief that v-like animals leave trees only by climbing down them (or that v-like
animals never fly, or some such). (Bennett 1976, 105)
In his earlier book Rationality Bennett argued that the interdependence
of basic inductive generalizations and conscious memories (in my terms)
meant that neither could be ascribed to nonlinguistic creatures. Only linguistic
creatures, he claimed, could manifest these two cognitive abilities
separately. Only in language can one manifest a conscious memory with-
184 Chapter 7
out a related inductive generalization, and vice versa. This yields his
restriction of these abilities to language users when coupled with the assumption
that a creature that cannot manifest two different types of cognitive
ability separately cannot properly be credited with either of them.8
As Bennett subsequently realized, however, the fact that basic inductive
generalizations and conscious memories always coexist in the explanation
of behavior if they feature in it at all does not mean that they cannot be
separated out on behavioral grounds. Suppose that we are faced with a
particular piece of behavior that appears to demand explanation in terms
of a conscious memory of a particular place and a concomitant inductive
generalization. Different inductive generalizations dictate different forms
of behavior in different circumstances. By examining what a creature does
in different circumstances, we can work out which inductive generalizations
are governing its behavior and apply this back to the original behavior.
Similarly, a given conscious memory of a place may, in conjunction
with different inductive generalizations, dictate identifiable forms of behavior
relative to that place. By comparing different candidate conscious
memories with different candidate inductive generalizations, it should
be possible in principle to narrow the field down to a single memorygeneralization
pair.
This interdependence of conscious memories and basic inductive generalizations
is very significant in dealing with the practical problem identified
earlier: how are we to establish the presence of explicit or conscious
memory in nonlinguistic creatures, given the importance of verbal reports
in deciding what memories to ascribe to linguistic creatures? Let me set
the scene by describing two sets of experiments in the animal learning
literature. The first is a set of experiments on monkeys (rhesus and pigtailed
macaques) carried out by Gaffan (1977). Monkeys were trained to
watch a screen on which 25 colored slides were presented twice at random
during a testing session. The monkeys were tested for their capacity
to detect whether a given slide was being presented for the first or second
time during a training session. Pressing a lever during the second presentation
of a slide was rewarded with a sugar pellet, but pressing the lever
during the first presentation was not. The two rhesus monkeys tested were
able (after training) to discriminate with 90 percent accuracy between
first and second presentation, although an average of 9 slides intervened
between the two presentations.