Current Biology 25, 256–260, January 19, 2015 ª2015 Elsevier Ltd All rights reserved
063
Crows Spontaneously Exhibit
Analogical Reasoning
Anna Smirnova,1 Zoya Zorina,1 Tanya Obozova,1
and Edward Wasserman2,*
1Department of Biology, Lomonosov Moscow State
University, Moscow 119899, Russia
2Department of Psychology, The University of Iowa, Iowa City,
IA 52242, USA
Summary
Analogical reasoning is vital to advanced cognition and
behavioral adaptation. Many theorists deem analogical
thinking to be uniquely human and to be foundational to
categorization, creative problem solving, and scientific dis-
covery [1]. Comparative psychologists have long been inter-
ested in the species generality of analogical reasoning, but
they initially found it difficult to obtain empirical support
for such thinking in nonhuman animals (for pioneering
efforts, see [2, 3]). Researchers have since mustered consid-
erable evidence and argument that relational matching-to-
sample (RMTS) effectively captures the essence of analogy,
in which the relevant logical arguments are presented visu-
ally [4]. In RMTS, choice of test pair BB would be correct if
the sample pair were AA, whereas choice of test pair EF
would be correct if the sample pair were CD. Critically, no
items in the correct test pair physically match items in the
sample pair, thus demanding that only relational sameness
or differentness is available to support accurate choice re-
sponding. Initial evidence suggested that only humans and
apes can successfully learn RMTS with pairs of sample
and test items [4–7]; however, monkeys have subsequently
done so [8–12]. Here, we report that crows too exhibit rela-
tional matching behavior. Even more importantly, crows
spontaneously display relational responding without ever
having been trained on RMTS; they had only been trained
on identity matching-to-sample (IMTS). Such robust and
uninstructed relational matching behavior represents the
most convincing evidence yet of analogical reasoning in a
nonprimate species, as apes alone [7] have spontaneously
exhibited RMTS behavior after only IMTS training.
Results and Discussion
Our findings come from three separate behavioral assess-
ments that followed pretraining and pretesting on identity
matching-to-sample (IMTS; see the Supplemental Experi-
mental Procedures available online), deploying behavioral
methods that were earlier used by Smirnova, Lazareva, and
Zorina [13]. In that prior IMTS period, two hooded crows
were shown several different kinds of visual stimuli: single
stimuli of varying colors or shapes and clusters of stimuli
containing varying numbers of items. The birds were taught
to match stimuli by color, shape, and number of items,
but not size. After learning these various IMTS tasks, the
crows reliably transferred discriminative responding to new
stimuli from the same categories that were used in training
*Correspondence: ed-wasserman@uiowa.edu
http://dx.doi.org/10.1016/j.cub.2014.11.
Report
(Table S2) as well as to stimuli from an altogether different
category (Figure S2)—stimuli varying in size—suggesting
that the birds had acquired a general rule that was based on
physical identity [14].
The key behavioral assessments followed, in which novel
pairs of items served as the sample and test stimuli [7]. The
visual dimensions in these assessments were size, shape,
and color (Figures 1, 2, and 3 show several illustrative, not
actual, trial sequences). On identity trials, we arranged differ-
ential reinforcement (food was given only after correct
choices) to promote continued discriminative responding;
accurate choice responding here could be based on either
physical or relational matches between the sample pair and
the correct test pair. On the critical relational trials, we
arranged nondifferential reinforcement (food was given after
all choices) because we did not want to teach the crows the
very behavior that we were assessing; accurate choice re-
sponding here could only be based on relational matches be-
tween the sample pair and the correct test pair. Assessment
sessions were conducted 5 days a week and contained 48
trials: 36 identity trials and 12 relational trials. The sessions
contained six blocks (each block containing six identity trials
and two relational trials); the trial order was randomized within
each of the six blocks. On all trials, we scored as ‘‘correct’’
those choices that accorded with relational matching. Each
assessment phase lasted eight sessions.
The first assessment phase involved size. Figure 1 shows
that on half of the trials the sample pair involved shapes of
the same size, whereas on the other half of the trials the sam-
ple pair involved shapes of different sizes. With the different-
sized sample and choice pairs, the large shape always
appeared on the left and the small shape always appeared
on the right in order to keep the number of sample-compari-
son combinations within manageable limits. The palette of
simple black shapes from which the sample and test pairs
could be selected contained a square, a rectangle, a circle,
an oval, a wide triangle, and a narrow triangle. These shapes
could be large or small; the overall area of the two test pairs
was equated, so that this stimulus property could not control
the crows’ behavior. Identity matching trials were arranged in
which one test pair presented the same shapes in the same
sizes as the sample pair; each of the sample and test pairs
involved the same shape, and only correct choices were
reinforced on these trials. Relational matching trials were
arranged in which neither of the test pairs matched the sample
pair in shape, thereby eliminating control by physical identity;
on these trials, either correct or incorrect choices were
reinforced.
The second assessment phase involved shape. Figure 2
shows that on half of the trials the sample pair involved iden-
tical shapes, whereas on the other half of the trials the sample
pair involved nonidentical shapes. In this assessment phase
and the next, the items in each pair were arranged along the
positive diagonal in order for the spatial arrangement to differ
from the horizontal arrangement in the prior assessment phase
and to keep the number of sample-comparison combinations
within manageable limits. The palette of shapes contained
equal-sized circles, squares, triangles, and crosses that could
be colored red, yellow, blue, or green. Identity matching trials
257

Trial Left Test Sample Right Test Trial Left Test Sample Right Test

1 1

2 2

3 3

4 4

5 5

6 6

7 7

8 8

Figure 1. Examples of Identity and Relational Trials across Eight Exemplary
Trials for Size Stimuli
On three-fourths of the trials (1–3 and 5–7), the correct test stimulus was an
identity match to the sample. On one-fourth of the trials (4 and 8, shaded
gray), the correct test stimulus was a relational match to the sample.
Figure 2. Examples of Identity and Relational Trials across Eight Exemplary
Trials for Shape Stimuli
On three-fourths of the trials (1–3 and 5–7), the correct test stimulus was
an identity match to the sample in shape. On one-fourth of the trials
(4 and 8, shaded gray), the correct test stimulus was a relational match to
the sample.

were arranged in which one test pair presented the same

shapes as the sample pair; only correct choices were rein-
forced on these trials. Relational matching trials were arranged
in which neither of the test pairs matched the sample pair
in shape or color, thereby eliminating control by physical iden-
tity; on these trials, either correct or incorrect choices were
reinforced.
The third assessment phase involved color (crows have
excellent color vision [15]). Figure 3 shows that on half of
the trials the sample pair involved identical colors, whereas
on the other half of the trials the sample pair involved
nonidentical colors. The palette of colors contained equal-
sized circles, squares, triangles, and crosses that could be
shown in red, yellow, blue, or green. Identity matching trials
were arranged in which one test pair presented the same
colors as the sample pair; only correct choices were rein-
forced on these trials. Relational matching trials were ar-
ranged in which neither of the test pairs matched the sample
pair in color or shape, thereby eliminating control by physical
identity; on these trials, either correct or incorrect choices
were reinforced.
In the size-assessment phase, 60 combinations of stimuli
were used as samples on relational trials (they were never
repeated within sessions and were presented a maximum
of two times, in different locations, across sessions) and 12
combinations of stimuli were used as samples on identity tri-
als (they were never repeated within sessions and were pre-
sented a maximum of 24 times across sessions). In the shape
and color assessment phases, trial-unique combinations of
stimuli were used as samples on all relational and identity
trials.
Throughout all three assessment phases, the crows ex-
hibited highly accurate choice responding on both identity
and relational trials (fourth column in Table 1). We individually
assessed each crow’s behavior with a logistic regression
examining trial type, stimulus dimension, and session. These
regression analyses revealed no significant changes in choice
accuracy over the eight sessions in each phase [crow 1: c2(1) =
0.94, p = 0.33; crow 2: c2(1) = 1.57, p = 0.21] and no interactions
of Session with the other variables (daily accuracy scores are
reported in Table S1, and details of the statistical analyses
are reported in the Supplemental Experimental Procedures);
therefore, subsequent discussion does not consider this
factor.
We next compared accuracy scores on the different kinds
of trials against 50% (representing random choice between
the test pairs). Averaged across all eight sessions, crow 1
responded at significantly above chance accuracy levels to
all six different kinds of stimuli (B = 1.06, SE = 0.068, Z =
15.57, p < 0.0001), with accuracy ranging from 61.46% to
79.51% correct. There was a significant main effect of trial
type [c2(1) = 3.99, p = 0.0457], with slightly lower accuracy
on relational trials (69.44%) than on identity trials (75.46%).
There was also a marginally significant trial type 3 stimulus
dimension interaction [c2(2) = 5.67, p = 0.059]. Follow-up tests
within each stimulus dimension disclosed that the interaction
was due to a significant effect of trial type for the color dimen-
sion (B = 20.68, SE = 0.25, Z = 2.74, p = 0.0061), but not for the
shape (B = 0.18, SE = 0.27, Z = 0.68, p = 0.49) or size dimen-
sions (B = 20.36, SE = 0.27, Z = 1.33, p = 0.183), suggesting
that accuracy on color relational trials was a bit poorer than
on the remaining kinds of trials. Averaged across all eight ses-
sions, crow 2 responded at significantly above chance accu-
racy levels to all six different kinds of stimuli (B = 1.13, SE =
0.067, Z = 16.34, p < 0.0001). Accuracy ranged from 71.18%
to 78.82% correct, with accuracy being numerically (but not
reliably) lower on identity trials (72.22%) than on relational tri-
als (80.56%). The logistic regression yielded no other signifi-
cant effects.
Because of the importance of spontaneity to interpreting the
crows’ relational matching-to-sample (RMTS) behavior, we
258

Table 1. Mean Percentage of Correct Choices by Crows 1 and 2

Trial Left Test Sample Right Test Dimension Trial Type Session 1 Sessions 1–8
Crow 1

1 Size Identity 77.78 79.51

relational 75.00 72.92
Shape identity 75.00 70.83
2 relational 91.67 73.96
Color identity 66.67 76.04
3 relational 58.33 61.46
Crow 2
4 Size identity 75.00 76.39
relational 91.67 76.04
5 Shape identity 63.89 71.18
relational 75.00 77.08
Color identity 77.78 78.82
6 relational 75.00 71.88
Accuracy scores are from identity (IMTS) and relational (RMTS) trials on all
7 three dimensions in session 1 and across sessions 1–8. See also Table S1.

8
Figure 3. Examples of Identity and Relational Trials across Eight Exemplary
Trials for Color Stimuli
On three-fourths of the trials (1–3 and 5–7), the correct test stimulus was
an identity match to the sample in color. On one-fourth of the trials (4 and
8, shaded gray), the correct test stimulus was a relational match to the
sample.
also analyzed choice behavior in session 1 of each assess-
ment phase (third column in Table 1) using a similar logistic
regression as in the prior analyses. In session 1, crow 1 re-
sponded at significantly above chance levels of accuracy
(B = 1.08, SE = 0.20, Z = 5.39, p < 0.0001), but there were no sig-
nificant effects of trial type or stimulus dimension; summed
across all three dimensions, accuracy on identity trials aver-
aged 73.15% correct, and accuracy on relational trials aver-
aged 75.00% correct. Crow 2 also responded at significantly
above chance accuracy levels in session 1 (B = 1.11, SE =
0.20, Z = 5.53, p < 0.0001), but there were no significant effects
of trial type or stimulus dimension; summed across all three
dimensions, accuracy on identity trials averaged 72.22%
correct, and accuracy on relational trials averaged 80.56%
correct.
We thus found that, when tested during the size-assess-
ment phase immediately after initial IMTS training, our crows
not only responded discriminatively on identity matching
trials, but also did so on relational matching trials, all of
these trials for the first time involving two-item sample and
test stimuli. These results represent striking behavioral evi-
dence that the crows spontaneously perceived the relation
between relations without ever having been explicitly trained
to do so.
That initial size assessment entailed sample and compari-
son stimuli that were horizontally ordered large (left) to small
(right) with different-sized pairs. Could that incidental spatial
cue have affected the accuracy of the crows’ later choice
behavior? Evidently not. The crows continued to respond at
similarly high levels of accuracy on identity and relational trials
during the shape and color assessment phases, during which
the items in the sample and comparison pairs were diagonally
arranged.
This documentation of RMTS behavior is particularly note-
worthy because our crows exhibited discriminative relational
matching (averaging 77.78% correct) that was just as robust
as their identity matching (averaging 72.69%) in session 1
across all three assessment phases. Although physical iden-
tity could have guided the crows’ choice behavior on IMTS tri-
als, physical identity could not have done so on RMTS trials, as
no physical matches were possible between the sample pairs
and the correct test pairs. These results perhaps surprisingly
suggest that physical identity contributed little or nothing to
our crows’ testing performance; relational processing seems
to have been of prime importance to controlling the birds’
choice behavior.
Our results thus constitute unprecedented behavioral evi-
dence of analogical reasoning by a nonprimate animal. They
therefore add to growing research undermining the influential
claims of such famous philosophers as René Descartes and
John Locke that only humans are capable of abstract thought.
Relational reasoning—particularly appreciating the relation
between relations, as in analogies—can no longer be deemed
to be the unique pinnacle of human cognition.
It may be no accident that crows performed so impressively
in our study; they stand out among birds in their highly devel-
oped neuroanatomy [16, 17]. More generally, mounting evi-
dence indicates that although birds do not have a brain
structure that is homologous to the mammalian prefrontal
cortex, the avian nidopallium caudolaterale may effectively
mediate complex cognitive functions, perhaps representing
a case of convergent evolution [18].
It should nevertheless be appreciated that we are not claim-
ing that our crows’ spontaneous relational matching behavior
arose entirely de novo. Indeed, we believe that their earlier
IMTS training is likely to have enabled them to grasp a broadly
applicable concept of sameness that could apply to novel two-
item sample and test stimuli involving only relational sameness
[19, 20]. Just how that remarkable transfer is accomplished
represents an intriguing matter for future study. Nor are we
claiming that crows will prove to be the only nonprimate ani-
mals that are capable of exhibiting such spontaneous rela-
tional matching behavior. Future research must be undertaken
in which different species are given comparable pretraining
experience to our crows. Until such systematic comparative
research is conducted, it would be premature to offer specula-
tive evolutionary accounts as to why crows appear to have
excelled in solving this challenging cognitive task. It would
also be premature to suggest that simply because crows
259
successfully master RMTS tasks, they process the stimuli and
relations in the same way as do humans and other nonhuman
animals.
Finally, we should underscore the importance of research on
animal behavior to the role of language or symbol systems in
abstract conceptualization. The first evidence of analogical
reasoning in animals came from symbol-trained chimpanzees
[2, 5]. Accordingly, the hypothesis was advanced that only if an
organism had acquired an elaborate symbol system might it
encode and process abstract analogical relations [5]. The re-
sults of more recent primate studies [6, 7, 12], as well as the
findings from our present experiment with crows, suggest
that extensive prior experience with the abstract concept of
‘‘sameness’’ per se—rather than a linguistic symbol for such
a concept—is sufficient for animals to succeed in solving
analogical reasoning tasks.
Experimental Procedures
Subjects
Two experimentally naive hooded crows (Corvus corone), at least 2 years
old, served as subjects. Both were housed in the aviary of the Biology
Department of Lomonosov Moscow State University, Russia. Throughout
the experiment, the birds had free access to water. Mealworms were
used as reinforcement, as they are crows’ favorite food and are attractive
at most food-deprivation levels. If the crows refused to work in the exper-
iment, then they received food without animal protein for 1 or 2 days.
Refusals to work meant that unfinished sessions were continued on the
next day. All of this research was conducted in full compliance with
Russian research regulations: specifically, the bioethical requirements of
Directive 86 EC.
Apparatus
A wire mesh cage (70 cm 3 35 cm 3 35 cm; 4 cm 3 4 cm mesh spacing) and
a plastic tray (20 cm 3 30 cm) with a handle (30 cm) were used for training
and testing (Figure S1). Two cups (3.7 cm high and 5.0 cm in diameter)
were placed on the tray; during training, one cup contained two mealworms
and the other cup was empty, whereas during testing both cups contained
mealworms. The cups were covered by the comparison stimuli. The sample
stimulus was placed between the comparison stimuli. All of the stimuli were
drawn on cardboard cards (7 cm 3 7 cm). An opaque plastic screen (70 3
40 cm) was placed between the experimenter and the crow; neither the
bird nor the experimenter could see one another, precluding a ‘‘Clever
Hans’’ error. Before each trial, the tray was prepared out of the bird’s sight.
The Supplemental Information includes additional procedural details, con-
trols for confounded variables, and Movie S1 (a reenactment showing
crow 1 performing the task).
General Procedure
The crows were trained and tested in two-alternative simultaneous match-
ing-to-sample tasks. During the experiment, a bird was placed into the
experimental cage. A trial started when the tray—containing the sample
stimulus card in the center and the two cups covered by the comparison
stimulus cards on each side—was slid into the cage. So that the crow could
be given the opportunity to get acquainted with all three stimuli, the tray was
initially placed in front of the bird for 2 to 3 s as the first step of the trial; the
crow could see the cards, but it could not uncover the cups. Then, the sec-
ond step followed. The tray was moved more deeply into the cage; the crow
uncovered one of the cups and, in case of the correct choice, it received
food. If the bird did not choose either card within 2 min, then the tray was
removed from the cage. Which sample stimulus would be presented was
determined by a quasirandom schedule under the restrictions that (1) the
same card could not be used as a sample more than two times in succession
and (2) the correct stimulus could not appear in the right or left location more
than two times in succession. Otherwise, the order of stimulus presentation
and the pairing of sample and comparison stimuli were randomized on each
trial.
The minimum intertrial interval (ITI) was about 1 min, essentially the time
to prepare the tray with new stimuli and mealworms. The maximum ITI was
5 min. If the bird did not choose either card within 2 min, then the tray was
removed from the cage. The experimenter gave the bird a 5 min break and
repeated that trial. If the bird again did not choose either card, then the
experimenter ended the session.
Supplemental Information
Supplemental Information includes Supplemental Experimental Proce-
dures, two figures, two tables, and one movie and can be found with this
article online at http://dx.doi.org/10.1016/j.cub.2014.11.063.
Author Contributions
A.S. and Z.Z. designed and conducted the studies and interpreted the data.
T.O. and E.W. helped conduct the statistical analyses and interpret the data
and composed the paper in collaboration with the other coauthors.
Acknowledgments
This research was supported by Russian Foundation for Basic Research
(RFBR) grant 13-04-00747. We thank Leyre Castro, Bob McMurray, and
Mark Blumberg for their help in preparing this report.
Received: September 8, 2014
Revised: November 3, 2014
Accepted: November 26, 2014
Published: December 18, 2014
References
1. Gentner, D. (1999). Analogy. In The MIT Encyclopedia of the Cognitive
Sciences, R.A. Wilson and F.C. Keil, eds. (Cambridge: MIT Press),
pp. 17–20.
2. Gillan, D.D., Premack, D., and Woodruff, G. (1981). Reasoning in the
chimpanzee: I. Analogical reasoning. J. Exp. Psychol. Anim. Behav.
Process. 7, 1–17.
3. Oden, D.L., Thompson, R.K.R., and Premack, D. (2001). Can an ape
reason analogically? Comprehension and production of analogical
problems by Sarah, a chimpanzee (Pan troglodytes). In The Analogical
Mind: Perspectives from Cognitive Science, D. Gentner, ed.
(Cambridge: MIT Press).
4. Thompson, R.K.R., and Oden, D.L. (2000). Categorical perception and
conceptual judgments by nonhuman primates: The paleological mon-
key and the analogical ape. Cogn. Sci. 24, 363–396.
5. Premack, D. (1983). The codes of man and beast. Behav. Brain Sci. 6,
125–137.
6. Thompson, R.K.R., Oden, D.L., and Boysen, S.T. (1997). Language-
naive chimpanzees (Pan troglodytes) judge relations between relations
in a conceptual matching-to-sample task. J. Exp. Psychol. Anim. Behav.
Process. 23, 31–43.
7. Vonk, J. (2003). Gorilla ( Gorilla gorilla gorilla) and orangutan ( Pongo
abelii) understanding of first- and second-order relations. Anim. Cogn.
6, 77–86.
8. Fagot, J., and Maugard, A. (2013). Analogical reasoning in baboons
(Papio papio): flexible reencoding of the source relation depending on
the target relation. Learn. Behav. 41, 229–237.
9. Fagot, J., and Parron, C. (2010). Relational matching in baboons (Papio
papio) with reduced grouping requirements. J. Exp. Psychol. Anim.
Behav. Process. 36, 184–193.
10. Fagot, J., and Thompson, R.K.R. (2011). Generalized relational match-
ing by guinea baboons (Papio papio) in two-by-two-item analogy prob-
lems. Psychol. Sci. 22, 1304–1309.
11. Flemming, T.M., Thompson, R.K.R., and Fagot, J. (2013). Baboons, like
humans, solve analogy by categorical abstraction of relations. Anim.
Cogn. 16, 519–524.
12. Fagot, J., Wasserman, E.A., and Young, M.E. (2001). Discriminating the
relation between relations: the role of entropy in abstract conceptualiza-
tion by baboons (Papio papio) and humans (Homo sapiens). J. Exp.
Psychol. Anim. Behav. Process. 27, 316–328.
13. Smirnova, A.A., Lazareva, O.F., and Zorina, Z.A. (2000). Use of number
by crows: investigation by matching and oddity learning. J. Exp. Anal.
Behav. 73, 163–176.
14. Katz, J.S., Wright, A.A., and Bodily, K.D. (2007). Issues in the compara-
tive cognition of abstract-concept learning. Comp. Cogn. Behav. Rev. 2,
79–92.
260

15. Håstad, O., Victorsson, J., and Ödeen, A. (2005). Differences in color
vision make passerines less conspicuous in the eyes of their predators.
Proc. Natl. Acad. Sci. USA 102, 6391–6394.
16. Emery, N.J. (2006). Cognitive ornithology: the evolution of avian intelli-
gence. Philos. Trans. R. Soc. Lond. B Biol. Sci. 361, 23–43.
17. Portmann, A. (1947). Études sur la cérébralisation chez les oiseaux. II.
Les indices intracérébraux. Alauda 15, 1–15.
18. Güntürkün, O. (2012). The convergent evolution of neural substrates for
cognition. Psychol. Res. 76, 212–219.
19. Truppa, V., Piano Mortari, E., Garofoli, D., Privitera, S., and Visalberghi,
E. (2011). Same/different concept learning by capuchin monkeys in
matching-to-sample tasks. PLoS ONE 6, e23809.
20. Wasserman, E.A. (2008). Development and evolution of cognition: one
doth not fly into flying!. Behav. Brain Sci. 31, 400–401.
Current Biology, Volume 25
Supplemental Information

Crows Spontaneously Exhibit

Analogical Reasoning
Anna Smirnova, Zoya Zorina, Tanya Obozova, and Edward Wasserman
Figure S1. Photograph of the experimental apparatus showing the location of the experimenter

and the subject when the crow made its choice.

 Trial Left Test Sample Right Test

Figure S2. Examples of sample and comparison stimuli across 8 exemplary IMTS testing trials

outside the training categories. On three-fourths of the trials (1-3 and 5-7), only correct choices

were reinforced; these were trials where the sample stimulus was identical to one of the

comparison stimuli. On one-fourth of the trials (4 and 8, shaded grey), both correct and incorrect

choices were reinforced; these were trials where the sample stimulus was not identical in shape

to either of the comparison stimuli, but matched the comparison stimuli on the basis of the novel

category of size.
 Bird Dimension Trial Type Session 1 Session 2 Session 3 Session 4 Session 5 Session 6 Session 7 Session 8 Sessions 1 - 8

Identity 77.78 75.00 86.11 86.11 72.22 80.56 77.78 80.56 79.51
Size
Relational 75.00 83.33 83.33 66.67 66.67 50.00 75.00 83.33 72.92

Identity 75.00 80.56 61.11 69.44 80.56 69.44 63.88 66.67 70.83
Crow 1 Shape
Relational 91.67 75.00 83.33 66.67 75.00 66.67 66.67 66.67 73.96

Identity 66.67 80.56 86.11 75.00 69.44 75.00 63.88 91.67 76.04
Color
Relational 58.33 75.00 50.00 66.67 75.00 50.00 50.00 66.67 61.46
 Identity 75.00 72.22 72.22 66.67 86.11 72.22 86.11 80.56 76.39
Size
Relational 91.67 75.00 83.33 75.00 58.33 80.56 66.67 75.00 76.04

Identity 63.89 63.88 63.88 75.00 83.33 66.67 75.00 77.78 71.18
Crow 2 Shape
Relational 75.00 75.00 91.67 58.33 100.00 83.00 66.67 66.67 77.08

Identity 77.78 72.22 72.22 80.56 86.11 83.33 77.78 80.56 78.82
Color
Relational 75.00 66.67 66.67 75.00 75.00 83.33 58.33 75.00 71.88

Table S1. Mean percentage of correct choices by Crows 1 and 2. Accuracy scores are from Identity (IMTS) and Relational (RMTS) trials on

all three dimensions across Sessions 1-8.

 Trial Left Test Sample Right Test
1 white white black
2 array of 4 items array of 3 items array of 3 items
3 Light grey dark grey dark grey
4 array of 5 items array of 5 items array of 6 items
5 numeral 1 numeral 1 numeral 2
6 numeral 2 numeral 3 numeral 3
7 black black white
8 numeral 5 numeral 8 numeral 8

Table S2. Examples of sample and comparison stimuli across 8 exemplary trials of IMTS testing

within the training categories. On three-fourths of the trials (1-3 and 5-7) only correct choices

were reinforced; these were trials with the familiar training stimuli. On one-fourth of the trials (4

and 8, shaded grey) both correct and incorrect choices were reinforced; these were trials with the

novel testing stimuli from the training sets.

Supplemental Experimental Procedures

Pretraining and Pretesting Steps

The crows’ pretraining and pretesting was on IMTS tasks in the apparatus shown in

Figure S1. Here, the crows’ choice responses were reinforced for selecting the comparison

stimulus that was identical to the sample. It is important to stress that, in order to conclude that

crows had acquired a “matching” concept, one must provide clear evidence of successful

learning as well as successful transfer of matching behavior to novel stimuli. If the birds

accurately apply the matching rule to completely novel stimuli, then it may be concluded that the

crows’ performance is conceptually mediated.

We thus trained the crows to acquire the concept of sameness based on color, Arabic

numerals, and the number of items composing the stimuli. We later tested their ability to match

novel stimuli from the same categories and from an entirely new category—size.

IMTS Training

IMTS training continued until the acquisition criterion (80% correct choices over 96

consecutive trials; binomial test, p < .001) had been reached for each of four successively given

stimulus sets: (a) black and white, (b) Arabic numerals 1 and 2, (c) arrays of 1 and 2 elements,

and (d) intermixed pairs (black and white, light grey and dark grey, Arabic numerals from 1 to 4,

and arrays containing 1 to 4 elements).

To reach the criteria for each of four stimulus sets, Crow 1 needed 2,096, 1,056, 128, and

96 trials, respectively; Crow 2 needed 3,136, 128, 104, and 96 trials, respectively.

IMTS Testing within the training categories

After training, we explored whether the crows could transfer their IMTS behavior to

novel stimuli within the training categories. On three-fourth of the trials, all of the stimuli came

from the training set. On another one-fourth of the trials, none of the stimuli came from the

training set.
 Each crow completed 288 trials over 6 sessions. On each of the trials, the correct

comparison stimulus was an exact match to the sample. On 216 trials, all of the stimuli had

previously been seen during training. On another 72 trials, none of the stimuli had previously

been seen during training. The testing stimuli came from the same categories: four different

colors, Arabic numerals from 5 to 8, and arrays containing 5 to 8 elements. [A series of 8

exemplary trials is presented in Table S2.]

Binomial tests for choice accuracy on both familiar stimulus training trials and novel

stimulus testing trials revealed that each crow performed at significantly above chance levels

(50%), p < 0.001, thereby attesting to the ability of the birds to transfer a matching rule to new

stimuli from the same training category. The mean percentage of correct choices for Crow 1 was

81.48% on training trials and 83.33% on testing trials. The mean percentage of correct choices

for Crow 2 was 83.33% on training trials and 79.16% on testing trials.

IMTS Testing outside the training categories

This final step explored whether the crows could transfer their IMTS behavior to stimuli

from a novel category—size. On three-fourth of the trials, the correct comparison stimulus was

an exact match to the sample. On another one-fourth of the trials, the correct comparison

stimulus matched the sample in size, but depicted a different shape. Each testing stimulus

involved six different black shapes on a white background. Each shape appeared in one of two

sizes: large and small.

Both crows completed 480 trials over 10 sessions. On 360 trials, the correct comparison

stimulus was an exact match to the sample (exact match). On another 120 trials, the correct

comparison stimulus matched the size, but not the shape of the sample (size match). [A series of

8 exemplary trials is in Figure S2.] Because of the importance of assessing the immediacy of

transfer outside of the training categories, only the scores from Session 1 are reported here.
 Binomial tests for choice accuracy in Session 1 of testing revealed that each bird

consistently performed at above chance levels (50%), all p < 0.001, thereby documenting the

ability of crows to transfer IMTS to novel stimuli outside the training categories. The percentage

of correct choices for Crow 1 was 83.33% on exact match trials and 83.33% on size match trials.

The percentage of correct choices for Crow 2 was 88.89% on exact match trials and 91.67% on

size match trials. These scores indicate that the crows had indeed acquired a general matching

rule based on the abstract concept of sameness before the final step of RMTS testing had begun.

Controls for Confounded Visual and Auditory Stimuli

To control for any inadvertent visual cues, the stimulus tray was prepared for each trial

out of the field of the bird’s vision. An opaque plastic screen (70 × 40 cm) was placed between

the experimental cage and the experimenter, minimizing the chance of unintentional cues from

the experimenter. The bird could not see the experimenter at the moment of choice nor could the

experimenter see the bird. This methodological measure precluded the possibility of a “Clever

Hans” error. Note that because the experimenter could not see the bird, she could only judge the

outcome of each trial by the sounds made by the bird. If she heard the sound of a card falling

from the cup and the later sound of the bird pecking the cup containing the worms, then the

choice was correct. If no pecking sound occurred, then it meant that the choice was incorrect and

the experimenter quickly removed the tray to prevent the bird from uncovering the second cup.

To control for any inadvertent auditory cues, the tray was always prepared in the same

sequence for each trial: the left cup was placed on the tray, then right cup was placed on the tray.

One of the cups contained two mealworms as reinforcement. Subsequently, the left cup was

covered with a comparison stimulus, then right cup was covered with a comparison stimulus.

The mealworms were alive and may have emitted sounds, but the importance of such sounds

could be eliminated on RMTS trials because mealworms were placed into each cup.
Statistical Analysis Details During Identity and Relational Assessment Phases

Our primary data were analyzed with logistic regressions run separately for each crow.

Each analysis examined accuracy as a function of three independent factors. Session (1-8) was a

linear predictor and centered. Trial Type (Identity vs. Relational) was dummy coded (Relational

= 1) and then centered. Stimulus Dimension was coded as two dummy variables: one was set to

1 for Size and 0 otherwise; the other was set to 1 for Color and 0 otherwise. Both were then

centered. Because the main effect of Stimulus Dimension (and its various interactions) was

spread across two variables, the significance of fixed effects was established with the χ2 of

model comparison. Main effects were evaluated by comparing a model with all three main

effects to one without the effect in question. Two-way interactions were evaluated by comparing

a model with all two-way interactions to one without the interaction in question. The three-way

interaction was evaluated by adding the three-way interaction terms to a model with only two-

way interactions. The intercept of the model (and its significance) tests the hypothesis that

accuracy is above (or below) chance; we report the Wald Z statistic for this purpose. Models

were run using the GLM command (family = “binomial”) in R (Version 3.1.0).