Work with Nonhuman Animals
Our research on abstract categorization in the pigeon has
focused on the acquisition and nature of the same-different concept.
We have identified variability as an important underlying dimension
to this concept, demonstrated the generality of the concept,
and are currently conducting further research into the nature
of this concept. We are not alone, however, in our endeavors
to understand the characteristics of the same-different concept
in nonhuman species.
Research into the same-different concept in pigeons began
with investigations using matching-to-sample (MTS) and non-matching-to-sample
(NMTS) tasks (reviewed by Delius, 1994). In these tasks, the
pigeon is first given a sample stimulus (e.g., a red light) and
subsequently given a choice of two stimuli (e.g., a red light
or a green light). In MTS, the pigeon must choose the stimulus
that matches the sample (in this case the red light), whereas
in NMTS the pigeon must choose the stimulus that does not match
the sample (in this case the green light). Although it has been
clearly demonstrated that pigeons can perform well during transfer
tests (see Delius, 1994), MTS and NMTS tasks may not involve
an understanding of same-different at the conceptual level. In
fact, Premack (1976) has argued that strong evidence of the presence
of a concept of same-different would require that an observer
be able to assign arbitrary responses to items that were either
the same as or different from one another. Our work follows a
tradition of demonstrating that pigeons are intelligent enough
to meet Premack's criterion for abstract conceptual behavior.
Wright, et al. (1983) trained both pigeons and rhesus monkeys
to perform a same-different discrimination in which two identical
slides required one response and two different slides required
a second response. After considerable training, both of the monkeys
learned the discrimination and showed near perfect transfer to
novel item pairs. In contrast, none of the pigeons performed
well on the transfer tests. This observation led Wright et al.
(1983) to a rather pessimistic conclusion: "There is the
possibility that monkeys can abstract the task and learn the
Same/Different concept whereas pigeons cannot" (p. 316).
Our research corroborates the difficulty of teaching pigeons
to classify two items as same or different; but, we have further
demonstrated that they can learn the concept when eight or more
items are the same as or different from one another. Thus, the
difference between pigeons and monkeys noted by Wright et al.
(1983) appears to be a quantitative one and not a qualitative
Oddity detection. Cook
and colleagues (e.g., Cook, 2001; Cook, Cavoto, &
Cavoto, 1995; Cook, Katz, & Cavoto, 1997) have used a different
paradigm to investigate same-different processing in the pigeon.
They typically present pigeons with arrays containing a large
collection of elements (e.g., 384); the pigeons' task is to choose
one key ("Same") when the array is uniform (containing
identical elements) and a second key ("Different")
when a region of different elements (all identical to one another)
is present on a uniform background. Their pigeons have little
difficulty acquiring this discrimination, despite the fact that
the entropy differences (0.0 for Same displays, somewhat less
than 1.0 for Different displays) were smaller than that between
our 2-item Same and Different displays (0.0 and 1.0, respectively),
displays which our pigeons had great difficulty discriminating
(Young, Wasserman & Garner, 1997).
These results and others (see Cook et al., 1997) reveal that
pigeons learning our variability discrimination task and those
learning Cook and colleagues' oddity detection task may have
learned quite different rules. Whereas our task prompts discrimination
based on variability, their task may prompt a more oddity-based
evaluation of the same-different relations (see Cook et al.,
1997, for a similar conclusion).
Relative numerosity. Honig
and colleagues (1991; Honig & Matheson, 1995; Honig &
Stewart, 1989) have reported evidence which suggests that pigeons
are sensitive to the relative numerosity of two items arranged
in multi-element arrays. One of their experimental methods represents
a simpler version of the simultaneous same-different task that
we have used. Honig and Matheson (1995, Experiment 2) used a
uniform-mixture discrimination task that required the pigeons
to peck at one side of the screen for uniform arrays, consisting
of identically-colored squares, and the other side for mixture
arrays, consisting of various proportions of two colors of squares.
Honig (1991) and Honig and Matheson (1995) found that pigeons
could learn to discriminate between uniform arrays and all of
the various mixture arrays in the uniform-mixture task. In a
reanalysis of their data (Young & Wasserman, 1997) we discovered
that entropy, rather than relative numerosity, might be producing
the pigeons' discriminative performance in the uniform-mixture
Matching same-different relations.
Thompson, Oden, and Boysen (1997) have recently demonstrated
that language-naive chimpanzees can judge relations between relations
in a conceptual MTS task. This application of MTS differs from
that previously used in studies of the same-different concept.
The samples in Thompson et al.'s (1997) conceptual MTS problems
consisted of either two identical items (e.g., AA) or two different
items (e.g., AB). The chimpanzee was then required to choose
from two alternatives: one of which consisted of two identical
items (e.g., EE) and the other of which consisted of two different
items (e.g., FG); none of the items appearing in the sample were
members of the alternatives. This task thus required the chimpanzee
to match a relation - which of the alternatives involved the
same relation between its constituent items as that between the
constituent items of the sample - and is thus decidedly more
conceptual than the conventional MTS task. Thompson et al. (1997)
found that chimpanzees can learn this task and, furthermore,
that prior language training was not a prerequisite, although
experience with arbitrary tokens that were consistently associated
with abstract relations may be necessary.
Other species. In work
involving the more traditional simultaneous MTS or NMTS tasks,
ravens and gulls (Benjamini, 1983), and jackdaws, jays, and rooks
(Wilson, Mackintosh, & Boakes, 1985) have all been documented
to learn to choose the stimulus that matches the sample and to
generalize this learning to novel stimuli. Furthermore, Pepperberg
(1987) has trained a language-trained grey parrot (Alex) to identify
the specific difference between objects (matter, shape, or color).
Alex learned to properly identify the dimension on which the
objects differed at well above chance levels and to generalize
this knowledge to novel objects.
The bird's and the nonhuman primate's demonstrated abilities
to learn and use an abstract relational concept make one thing
clear: nonhuman animals are capable of conceptual feats heretofore
thought to be solely within the ken of the human species.
Related Work with
In research conducted in our human laboratory (Young &
Wasserman, 2001, in press), we were interested in quantifying
people's sensitivity to visual display variability and in determining
whether their discrimination of these displays would parallel
that of the pigeons used in our earlier studies. In our first experiment (Experiment
1, Young & Wasserman, 2001), we used the same simultaneous training procedure previously
used with pigeons. During a training phase, people were required
to discriminate Same from Different arrays. In tests involving
Mixture arrays of intermediate variability, we found that people's
predominant inclination was to treat same vs. different displays
categorically, not continuously. The majority of our participants
(80%) overwhelmingly chose the "different" key on Different
trials and Mixture trials (regardless of whether the Mixture
involved low variability or high variability); participants chose
the "same" key almost exclusively for the Same arrays.
Thus, similar tasks prompted two quite different patterns of
behavior in pigeons and in the majority of our human participants.
There were, nevertheless, some people (20% of our participants)
who did respond continuously to manipulations of display entropy.
In subsequent programmatic experiments, we sought to preclude
the participants' use of a categorical same-different distinction
so that the available contrast was in the degree of variability
present. Participants in these experiments were required to discriminate
among the Mixture arrays (each involving different levels of
variability). After learning to discriminate between low (Entropy
1.0) and high (Entropy 3.0) variability displays (Experiment
2, Young & Wasserman, 2001), people responded to displays of intermediate or more extreme
variability in close accord with their use of entropy as the
discriminative dimension. After learning to discriminate between
low (entropy less than 2.0) and high (entropy greater than 2.0)
variability displays (Young & Wasserman, in press), people again responded in
close accord with their use of entropy. Although people's use
of entropy was only revealed when a categorical distinction between
no variability and some variability was precluded, we did document
that people and pigeons both rely on entropy to discriminate
visual display variability.
The two species differed, however, in two fundamental ways.
First, most people were inclined to treat the discrimination
of intermediate degrees of variability as a categorical one (all
same vs. some different), whereas all pigeons treated the discrimination
as a continuous one. Second, although some training methods produced human variability discrimination
that was based on entropy, a significant number of college student
participants (and no pigeons) relied on relative entropy
rather than on absolute entropy.
Relative entropy is the absolute entropy of a display divided
by the maximum possible absolute entropy for a display comprising
the same number of items. Thus, any display containing items
that are all different from one another would have its maximum
possible absolute entropy (producing a relative entropy of 1.0).
The distinction between absolute and relative entropy is made
evident by examining the effect of the number of items in a display
on discriminative behavior. As previously reported (Young, Wasserman,
& Garner, 1997), pigeons were increasingly likely to choose
the "same" key as the number of items in a Different
display was reduced. The absolute entropy of these arrays also
decreases with the number of items. In contrast, relative entropy
is 1.0 for all of the Different arrays regardless of the number
of items comprising those displays. When tested with arrays involving
between 2 and 16 items, approximately half of our human participants
produced a response pattern consistent with a sensitivity to
absolute entropy and half produced a response pattern consistent
with a sensitivity to relative entropy.
Given the documented difficulty that pigeons have with discriminating
displays involving two identical items from those involving two
different items (a distinction between minimal and maximal relative
entropy), it is likely that pigeons find it very difficult to
compute entropy relative to some reference point (e.g., the maximal
entropy possible given the number of items present, as is required
to compute relative entropy).
Current and Future Directions
We have continued our investigations of same-different discrimination
on multiple fronts. In Wasserman, Young, and Nolan (2000)
we extended the notion of temporal organization (used in successive
same-different discrimination, Young et al., 1999) into the domain
of spatial organization in simultaneous same-different discrimination.
Although the spatial organization of the items did have a small,
but reliable effect on our pigeons' behavior, we demonstrated
that this effect can be accounted for by assuming that the pigeons
attended to a contiguous subset (approximately 6) of the 16 items.
In Young and Wasserman (2000), we have explored
the possibility that the pigeon's simultaneous same-different
discrimination might be driven by differences in perceptual regularities.
Analyses of our Same and Different
displays did in fact reveal a divergence in their perceptual regularity
(as indexed by their spatial frequency profile). In order to control for these differences in a new experiment,
we disrupted the regularities in Same arrays by jittering each
of the icons; each icon was offset both vertically and horizontally
by a small, random number of pixels. The jittering process successfully
produced nearly identical spatial frequency spectra for the Same
and Different arrays, but this manipulation had absolutely no
effect on our pigeons' discriminative performance. A second experiment revealed that
mixing orientations of the icons disrupted spatial regularity but not discriminative performance.
These results, in conjunction with the pigeon's ability to discriminate variability in lists of icons
(Young et al., 1999), strongly indicate that perceptual regularities are not the basis
of the pigeon's same-different discrimination.
Finally, we have begun a collaborative research effort with
Joel Fagot at the Center for Research in Cognitive Neuroscience
in Marseille, France in which baboons are being trained and tested
with a broad variety of simultaneous same-different discrimination
tasks. Baboons show rapid learning (less than
a week to reach criterion), strong transfer to novel items (only
a 10% difference between training and testing displays), and
continuous responding to same/different mixtures nearly identical
to that observed in pigeons (Wasserman, Fagot, & Young, in press). Given the observed differences
between the behavior of humans and pigeons in our tasks, the
examination of another primate species is helping to determine
the uniqueness of the human response profile.
Since its beginning a century ago, the science of comparative
cognition has sought to discover commonalities of mind between
humans and animals; it has also tried to provide a natural science
account of human and animal cognition (Wasserman, 1993; Wasserman,
1997). This movement now enters its second century equipped with
new and powerful methodological and technological tools for carrying
on its inquiry.
Our research represents one part of the current effort to
advance the science of comparative cognition. We combined a behavioral
analysis of concepts (Wasserman & Astley, 1994) with information
theory (Shannon & Weaver, 1949); used advanced computing
and video technology; and, accepted the challenge of cognitivists
to explore the most complex realms of memory and cognition instead
of complacently concentrating on the most elementary domains
of associative learning (Wasserman & Miller, 1997).
We are optimistic that this investigative strategy will continue
to disclose new and important parallels between human and animal
cognition, and that explanations for even the most advanced forms
of memory and cognition can be made exceedingly precise and rigorous.
A natural science of mind is not only possible, but necessary
if we are to move beyond the stale conventions that humans are
qualitatively different from animals and that mind is an inherently
inscrutable entity. Finally, showing that even pigeons are capable
of abstraction and that deterministic computational principles
may lie at the root of this complex cognitive process should
help stimulate neuroscientists to uncover the biological mechanisms
of this amazing ability.