Covariation of Learning and "Reasoning" Abilities in Mice: Evolutionary Conservation of the Operations of Intelligence

Abstract

Contemporary descriptions of human intelligence hold that this trait influences a broad range of cognitive abilities, including learning, attention, and reasoning. Like humans, individual genetically heterogeneous mice express a "general" cognitive trait that influences performance across a diverse array of learning and attentional tasks, and it has been suggested that this trait is qualitatively and structurally analogous to general intelligence in humans. However, the hallmark of human intelligence is the ability to use various forms of "reasoning" to support solutions to novel problems. Here, we find that genetically heterogeneous mice are capable of solving problems that are nominally indicative of inductive and deductive forms of reasoning, and that individuals' capacity for reasoning covaries with more general learning abilities. Mice were characterized for their general learning ability as determined by their aggregate performance (derived from principal component analysis) across a battery of five diverse learning tasks. These animals were then assessed on prototypic tests indicative of deductive reasoning (inferring the meaning of a novel item by exclusion, i.e., "fast mapping") and inductive reasoning (execution of an efficient search strategy in a binary decision tree). The animals exhibited systematic abilities on each of these nominal reasoning tasks that were predicted by their aggregate performance on the battery of learning tasks. These results suggest that the coregulation of reasoning and general learning performance in genetically heterogeneous mice form a core cognitive trait that is analogous to human intelligence.

Full text

Covariation of Learning and “Reasoning” Abilities in Mice: Evolutionary
Conservation of the Operations of Intelligence
Christopher Wass, Alexander Denman-Brice, Chris Rios, Kenneth R. Light, Stefan Kolata, Andrew M. Smith,
and Louis D. Matzel
Rutgers University
Contemporary descriptions of human intelligence hold that this trait influences a broad range of cognitive
abilities, including learning, attention, and reasoning. Like humans, individual genetically heterogeneous
mice express a “general” cognitive trait that influences performance across a diverse array of learning and
attentional tasks, and it has been suggested that this trait is qualitatively and structurally analogous to
general intelligence in humans. However, the hallmark of human intelligence is the ability to use various
forms of “reasoning” to support solutions to novel problems. Here, we find that genetically heteroge-
neous mice are capable of solving problems that are nominally indicative of inductive and deductive
forms of reasoning, and that individuals’ capacity for reasoning covaries with more general learning
abilities. Mice were characterized for their general learning ability as determined by their aggregate
performance (derived from principal component analysis) across a battery of five diverse learning tasks.
These animals were then assessed on prototypic tests indicative of deductive reasoning (inferring the
meaning of a novel item by exclusion, i.e., “fast mapping”) and inductive reasoning (execution of an
efficient search strategy in a binary decision tree). The animals exhibited systematic abilities on each of
these nominal reasoning tasks that were predicted by their aggregate performance on the battery of
learning tasks. These results suggest that the coregulation of reasoning and general learning performance
in genetically heterogeneous mice form a core cognitive trait that is analogous to human intelligence.
Keywords: intelligence, learning, reasoning, working memory, fast mapping, phylogenetics
Studies of individual differences in mental abilities in nonhu-
man animals have increased dramatically during the past decade,
and this work has served as an impetus to further delineate the
behavioral processes and neural mechanisms that underlie general
intelligence. Over a century ago, Spearman described “general
intelligence” (i.e., “g”), noting that a single factor could account
for a large portion of the underlying variance in performance
across diverse cognitive tasks. Spearman (1904) concluded that
regardless of the specific demands of a task, performance on that
task is determined by both domain-specific abilities (e.g., spatial
ability) as well as by g (i.e., general intelligence). It is the recog-
nition of this latent influence on the execution of most cognitive
tasks that underlies the rationale for contemporary tests of human
intelligence.
Despite decades of often vigorous debate, a single, a widely
accepted definition of “intelligence” has yet to be agreed upon.
Nevertheless, consensus has emerged regarding the operations
impacted by intelligence. In an article in the Wall Street Journal
(December 13, 1994) signed by 52 intelligence researchers, it was
asserted that intelligence was “a very general mental capability
that, among other things, involves the ability to reason, plan, solve
problems, think abstractly, comprehend complex ideas, learn
quickly and learn from experience.” A committee of the American
Psychological Association (1995) stated that “Individuals differ
from one another in their ability to understand complex ideas, to
adapt effectively to the environment, to learn from experience, and
to engage in various forms of reasoning to overcome obstacles.
Concepts of “intelligence” are attempts to clarify and organize this
complex set of phenomena.” While there is an apparent consensus
regarding the functional consequences of intelligent behavior,
these “definitions” are simultaneously nebulous in content and
expansive in scope.
While a widely accepted definition of “intelligence” has not
fully materialized, most descriptions (including those provided
above) state that intelligent behavior involves the ability to “adapt
to the environment” and to “learn quickly.” These characteristics,
like colloquial impressions of intelligence, suggest that learning is
a critical component of, if not critically influenced by, intelligence.
For instance, rates of learning among humans are often predicted
by psychometric tests of IQ (for reviews, see Gettinger, 1984;
Jensen, 1989). In fact, Jensen has concluded that general learning
factors (extracted from performance across large numbers of
learning tasks) are so highly related to the general intelligence
factor extracted from psychometric tests that “learning and psy-
chometric abilities (general intelligence) are essentially one and
the same” (Jensen, 1998). It is notable that occasional reports of a
Christopher Wass, Alexander Denman-Brice, Chris Rios, Kenneth R.
Light, Stefan Kolata, Andrew M. Smith, and Louis D. Matzel, Department
of Psychology, Rutgers University.
This work was supported by a grant from the U.S. Public Health Service
(National Institute of Aging, AG022698) to L.D.M.
Correspondence concerning this article should be addressed to Louis D.
Matzel, Program in Behavioral Neuroscience, Department of Psychology,
Rutgers University, Piscataway, NJ 08854. E-mail: matzel@rci.rutgers.edu
Journal of Experimental Psychology: © 2012 American Psychological Association
Animal Behavior Processes
2012, Vol. , No. , 000– 000
0097-7403/12/$12.00 DOI: 10.1037/a0027355
1
AQ: 1
AQ: 2
AQ: 3
AQ: 4
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

lack of correlation between performance on single learning tasks
and measures of general intelligence may at times be attributable
to the high task-specific variability inherent to measures of per-
formance on a single task, such that the influence of a general
factor on that task is relatively small (and thus, not detected by a
probability test). Furthermore, on elemental tasks that all subjects
can presumably master, any correlation between learning and
general intelligence is dependent on the measurement of learning
rate or speed of responding, a requirement that has at times been
overlooked (see Jensen, 1998, p. 275 for discussion relevant to
human learning, and Kolata, 2008, for a discussion relevant to the
present utilization of measures of learning performance; for alter-
native data and interpretation, see Williams & Pearlberg, 2006).
While the capacity for learning is indicative of intelligence,
definitions of intelligence (such as those provided above) make
reference to myriad processes beyond learning. Along with the
ability to learn, most definitions of this trait make primary refer-
ences to an individual’s capacity to “think rationally,” “reason,”
“engage in reasoning,” or colloquially, “to figure out novel solu-
tions” based on limited experience. Thus, the efficacy of reasoning
is widely considered to be a critical component of intelligence
(Manktelow, 1999), and most intelligence test batteries include
components specifically intended to characterize an individual’s
capacity for reasoning. In fact, many modern tests of intelligence
(e.g., the Raven’s Progressive Matrix) are weighted almost exclu-
sively toward this ability.
While it has been empirically determined that human intelli-
gence (including its expression on tests of reasoning) and learning
abilities covary (Gettinger, 1984; Jensen, 1989), there have been
no attempts to assess the relationship between general learning
abilities and reasoning in nonhuman animals. In this regard, it was
until recently a common assertion that reasoning was beyond the
capacity of nonhuman animals. In contrast to this assertion, dem-
onstrations of behaviors that are nominally indicative of reasoning
in nonhuman animals have begun to be reported. Briefly, species
such as monkeys (McGonigle & Chalmers, 1977; Rapp, Kansky,
& Eichenbaum, 1996), chimpanzees (Boysen, Berntson, Shreyer,
& Quigley, 1993), rats (Dusek & Eichenbaum, 1997; Roberts &
Phelps, 1994), and pigeons (Wynne, 1997; Lazareva, 2006 1853)
have all demonstrated the capacity for transitive inference, an
ability that, at least in some instances, has been asserted to depend
on deductive forms of reasoning (for alternative explanations, see
General Discussion & Shettleworth, 2010). Another task (that we
will adapt here for the assessment of mice) that has been used to
assess reasoning abilities in nonhuman animals is “fast mapping.”
Fast mapping is a mental process whereby a new concept can be
learned based on a logical inference derived from a single expo-
sure to limited information. Fast mapping (Care & Bartlett, 1978)
is believed to play a critical role in the extraordinarily rapid and
seemingly effortless acquisition of information during early human
development, and explains (in part) the prodigious rate at which
children gain vocabulary. For example, when faced with a group of
familiar items described by familiar words, an individual will
quickly associate an unfamiliar word with a novel item added to
the otherwise familiar set (“inference by exclusion”), and this
association requires no overt “pairing” of the novel word and its
corresponding novel item. Over time, the word’s approximate
meaning becomes more refined as it is seen in other contexts.
Logical inference, as exemplified here, is often described as a
hallmark of reasoning.
Although fast mapping has been extensively studied in humans,
there have been few attempts (except in two dogs and one chim-
panzee) to assess whether nonhuman animals are capable of this
cognitive process. Briefly, Kaminski, Call, and Fischer (2004),
demonstrated that a Border Collie (also see Pilley & Reid, 2011)
was able (on first command) to retrieve a novel object (identified
by a novel term) from among set of over 200 previously learned
objects. Kaminski et al. concluded that the Border Collie was able
to correctly retrieve the novel object through the use of inferential
exclusion principles, that is, the novel word used in the voice
command was inferred to mean to retrieve the novel object.
Hashiya and Kojima (2001) demonstrated that a chimpanzee was
also able to use inferential exclusion in order to correctly match a
novel voice to a novel portrait when two previously learned
portraits (distracters) were present. Based on the limited observa-
tions of fast mapping in nonhuman animals, it is reasonable to ask
whether mice are capable of utilizing similar basic reasoning, and
whether the fast mapping abilities of mice covary with the general
learning abilities of these animals.
Prior studies have demonstrated the existence of a general
cognitive factor in genetically heterogeneous mice, and this latent
influence on diverse learning abilities is in many ways analogous
to general intelligence in humans. Specifically, we have shown
that when genetically heterogeneous mice are assessed on batteries
of as many as nine diverse learning tasks, as much as 28 –50% of
the variance in performance across these tasks can be accounted
for by a single factor (Kolata et al., 2008; Kolata et al., 2010;
Light, Kolata, Denman-Brice, Zagalsky, & Matzel, 2008; Matzel
et al., 2003; Matzel et al., 2006; Matzel, Grossman, Light,
Townsend, & Kolata, 2008; also, see Galsworthy, Paya-Cano,
Monleo´n, & Plomin, 2002; Locurto, Fortin, & Sullivan, 2003).
Moreover, it has been determined that the efficacy of processing
components of working memory (i.e., capacity and selective at-
tention) covary with an animal’s general learning ability (Kolata et
al., 2005, 2007), and furthermore, working memory training reg-
imens promote an increase in animals’ general learning abilities
(Light et al., 2010; Matzel et al., 2011). These observations cor-
respond with studies of humans which have indicated that working
memory capacity covaries with an individual’s general intelligence
(Conway, Kane, & Engle, 2003; Engle, Laughlin, Tuholski, &
Conway, 1999; Jaeggi et al., 2009). Thus, the processes that
regulate the general learning abilities of mice are to some degree
homologous with those implicated in establishing general intelli-
gence in humans.
As reasoning abilities are widely considered to epitomize the
functional operations of “intelligence” (and comprise a core com-
ponent of all definitions of intelligence), the work reported in the
present article was aimed to determine whether laboratory mice are
capable of behaviors that are analogous to behaviors described as
“reasoning” among humans, and if so, whether the animals’ ca-
pacity for this reasoning-like behavior covaries with their general
learning abilities. To assess the hypothesis that general learning
abilities and reasoning-like cognitive performance covary in lab-
oratory mice, we first utilized a fast mapping-based task which is
commonly asserted to reflect a subject’s capacity for deductive
reasoning (e.g., Carey & Bartlett, 1978).
2WASS ET AL.
AQ: 5
AQ: 6
AQ: 7
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

Experiment 1A
In order to assess whether an animal’s general learning and
deductive reasoning abilities were jointly regulated, we first char-
acterized the general learning ability of individual animals in a
five-task learning battery that has been previously described (e.g.,
Matzel et al., 2003, 2006). A principal component analysis was
then applied to animals’ performance on these five learning tasks
in order to derive each animal’s factor score (a measure of each
animal’s aggregate performance across all learning tasks; see be-
low). Having ascertained the general learning ability of each
animal, we then assessed the same animals on a fast-mapping task.
Animals’ general learning performance was then compared to their
performance on the fast mapping task so as to determine the degree
of covariance between individuals’ capacity for deductive reason-
ing and general learning abilities.
Method
Subjects. A sample of 47 genetically heterogeneous male
CD-1 mice were obtained from Harlan Laboratories (Indianapolis,
IN) at 45 days of age. Owing to an experimenter error during fast
mapping training, six animals could not be tested in this task. Upon
arrival in our vivarium, animals were individually housed in stan-
dard shoe box cages and were allowed ad libitum access to food
and water. All animals were allowed to acclimate to the vivarium,
which was maintained on a 12 hour light/dark cycle, for a period
of two weeks. During this period, the animals were handled by the
experimenter for 90 sec/day in order to mitigate differential stress
responses to the experimenter.
Learning battery. To quantify animals’ general learning
abilities, we evaluated animals’ performance on a battery of five
diverse tasks that impinged on different domains of learning,
sensory/motor, and motivational systems. All of the animals were
tested on these five tasks in the following order: Lashley III Maze,
spatial water maze, passive avoidance, associative fear condition-
ing, and odor guided discrimination. Two days of rest intervened
between each successive task in the learning battery. For tasks
utilizing food reinforcers, animals were food deprived 48 hours
prior to training by allowing only 90 min access to food within 2
hours of the end of the light cycle.
Lashley III maze. The Lashley III maze consisted of a start
box, four interconnected alleys, and a goal box containing a food
reward. (For an illustration of the Maze, see Matzel et al., 2003.)
Efficient performance in the maze required that the animal make
five spatial alternations to reach the goal box. Over trials, the
errors (i.e., wrong turns or retracing) committed in the animal’s
approach to the goal box decreased. The maze (scaled for mice)
was constructed of black Plexiglas. Each alley measured 58 6
cm, and had 16 cm high walls. A goal box was (20 cm long) was
located 10 cm from the end of the last alley. A 2 cm diameter white
cup was located in the rear portion of the goal box, and 45 mg
BioServe (rodent grain) pellets served as reinforcers. Illumination
was 80 Lux at the floor of the maze. The maze was isolated behind
a shield of white Plexiglas to prevent the use of extramaze land-
mark cues.
Food-deprived animals were acclimated and trained on two
successive days. On the day prior to acclimation, all animals were
provided with three BioServe food pellets in their home cages to
familiarize them with the novel reinforcer. On the acclimation day,
each mouse was placed in the four alleys of the maze, but the
openings between the alleys were blocked (with black Plexiglas)
so that the animals could not navigate between them. Each animal
was confined to the start and subsequent two alleys for 4 min, and
for 6 min in the last (goal) alley, where three food pellets were
present in the food cup. This acclimation period promotes stable
and high levels of activity on the subsequent training day. On the
training day, each animal was placed in the start box and allowed
to traverse the maze until it reached the goal box and consumed the
single food pellet present in the cup. Upon consuming the food, the
animal was returned to its home cage for a 20 min intertrial
interval (intertribal interval, ITI), during which the apparatus was
cleaned. After the ITI, the mouse was returned to the start box to
begin the next trial, and the sequence was repeated for five trials.
Both the latency and errors (i.e., a turn in an incorrect direction,
including those which result in path retracing) to enter the goal box
were recorded on each trial.
Spatial water maze. Animals were immersed in a round pool
of opaque water from which they were able to escape onto a
hidden (i.e., submerged) platform. The latency for animals to find
the platform decreased across successive trials. In this task, per-
formance of animals can improve across trials despite the animals
beginning each trial from a new start location. Such a procedure
mitigates egocentric navigation and promotes the animals’ depen-
dence on extramaze spatial landmarks. Typically, rodents‘ perfor-
mance in the water maze does not rely on fixed motor patterns (i.e.,
performance improves despite the animal’s irregular starting loca-
tion) or the presence of discernable cues within the maze (e.g.,
visual, tactile, or olfactory signals). Instead, performance is depen-
dent on the stability of extramaze cues, or “landmarks,” and is said
to reflect the animals’ representation of its environment as a
“cognitive map.”
We employ a protocol in which mice exhibit significant reduc-
tions in their latency to locate the escape platform within 5–10
training trials. To support rapid acquisition, animals were confined
in a clear Plexiglas cylinder on the safe platform for 6 min on the
day prior to training. Second, a considerably longer ITI (10 min)
was used than is typical (cf., 90 sec). Lastly, the maze, surround,
and water were black and visual cues were constructed of patterns
of lights.
A round black pool (140 cm diameter, 56 cm deep) was filled to
within 24 cm of the top with water made opaque by the addition of
nontoxic, water soluble, black paint. A hidden 11 cm diameter
black platform was in a fixed location 1.5 cm below the surface of
the water midway between the center and perimeter of the pool.
The pool was enclosed in a ceiling-high black curtain on which
five different shapes (landmark cues) were variously positioned at
heights (relative to water surface) ranging from 24 –150 cm. Four
of these shapes were constructed of strings of white LEDs (spaced
at 2.5 cm intervals) and included an “X” (66 cm arms crossing at
angles 40o from the pool surface), a vertical “spiral” (80 cm long,
7 cm diameter, 11 cm revolutions), a vertical line (31 cm) and a
horizontal line (31 cm). The fifth cue was constructed of two
adjacent 7 W light bulbs (each 4 cm diameter). A video camera
was mounted 180 cm above the center of the water surface. These
cues provided the only illumination of the maze, totaling 172 Lux
at the water surface.
3
REASONING AND GENERAL LEARNING
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

On the day prior to training, each animal was confined to the
escape platform for 360 sec.
Training was conducted on the two subsequent days. On Day 1
of training, animals were started from one of three unique locations
on each of six trials. (The pool was conceptually divided into four
quadrants, and one starting point was located in each of the three
quadrants that did not contain the escape platform. The starting
point on each trial alternated between the three available quad-
rants.) An animal was judged to have escaped from the water (i.e.,
located the platform) at the moment at which four paws were
situated on the platform, provided that the animal remained on the
platform for at least 5 sec. Each animal was left on the platform for
a total of 20 sec, after which the trial was terminated. Trials were
spaced at 10 min intervals, during which time the animals were
held in their home cages. On each trial, a 90 sec limit on swimming
was imposed, at which time any animal that had not located the
escape platform was placed by the experimenter onto the platform,
where it remained for 20 sec.
Animals were observed from a remote (outside of the pool’s
enclosure) video monitor, and animals’ performance was recorded
on video tape for subsequent analysis. Day 2 of training proceeded
as did Day 1, albeit with only four trials. After the last training
trial, a 90 min retention period began, after which animals were
tested with a “probe” trial. On the probe test, the escape platform
was removed from the pool, and all animals were started from
the first position for that day. A 90 sec test was conducted in
which the animals’ time searching in the target quadrant (where
the escape platform was previously located) and nontarget
quadrants were recorded.
One-trial passive avoidance. Animals learn to suppress
movement to avoid contact with aversive stimuli. This “passive
avoidance” response is exemplified in step-down avoidance pro-
cedures, where an animal is placed on a platform, whereupon
stepping off of the platform it encounters a footshock. Following
just a single encounter with shock, animals are subsequently re-
luctant to step off of the safe platform. The animals’ reluctance to
leave the platform is believed to not reflect fear, because typical
fear responses are not expressed in animals engaged in the avoid-
ance response. Upon stepping off the platform, animals here were
exposed to a compound of bright light and loud oscillating noise
rather than shock, so as not to duplicate stimuli between tasks (see
fear conditioning, below). Like more common procedures, our
variant of this task supports learning after only a single trial (i.e.,
subsequent step-down latencies are markedly increased).
A chamber illuminated by dim (50 Lux) red light was used for
training and testing. Animals were confined to circular (“safe”)
chamber (10 cm diameter, 8 cm high). The walls and floor of this
chamber were white, and the ceiling was translucent orange. The
floor was comprised of plastic rods (2 mm diameter) arranged to
form a pattern of 1 cm square grids. A clear exit door (3 CM
square) was flush with the floor of the safe compartment, and the
door was able to slide horizontally to open or close the compart-
ment. The bottom of the exit door was located 4 cm above the floor
of a second circular chamber (20 cm diameter, 12 cm high). This
“unsafe” chamber had a clear ceiling and a floor comprised of 4
mm wide aluminum planks that formed a pattern of 1.5 cm square
grids oriented at a 45
o
angle relative to the grids in the safe
compartment. When an animal stepped from the safe compartment
through the exit door onto the floor of the unsafe compartment, a
compound aversive stimulus comprised of a bright (550 Lux)
white light and “siren” (58 dBc above the 50 dB background,
2.4 –3.7 kHz; Radio Shack sound oscillator, Model 273– 057) was
initiated.
Animals were placed on the platform behind the closed exit
door. After 4 min of confinement, the door was retracted and the
latency of the animal to leave the platform and make contact with
the grid floor was recorded. Prior to training, step-down latencies
typically range from 8 –20 sec. Upon contact with the floor, the
aversive stimulus (light, noise, and vibration) was presented for 3
sec. Upon initiation of the aversive stimulus, mice retract onto the
safe chamber, at which time the platform door was closed and they
were again confined for 4 min. At the end of this interval, the door
was opened and the latency of the animal to exit the platform and
step onto the grid floor (with no aversive stimulation) was again
recorded. The ratio of posttraining to pretraining step-down laten-
cies was calculated for each animal and served to index learning.
It has previously been determined that asymptotic performance is
apparent in group averages following 2–3 training trials; thus,
performance after a single trial reflects, in most instances, subas-
ymptotic learning.
Associative fear conditioning. In this task the animals re-
ceived a tone (conditional stimulus, CS) paired with a mild foot-
shock (US). Two distinct experimental chambers were used (a
training context and a novel context). Each box was contained
within a sound and light-attenuating chamber. The training box
(16.5 26.5 20 cm) was brightly lit (50 Lux) with a clear
Plexiglas front/back, and one stainless steel and one clear Plexiglas
side wall. The floor was composed of a steel grid (5 mm spacing)
from which a 0.6 mA constant current footshock could be deliv-
ered from a shock scrambler (Lafayette Instruments, Lafayette,
IN). The novel chamber (23 21.5 19 cm) was dimly lit (4
Lux) and all of the walls and the floor were composed of clear
Plexiglas. In both boxes the tone (60 dB, 2.9 kHz) was delivered
by a piezoelectric buzzer (Med Associates, EV-203a).
The animals were acclimated to the training and novel contexts
by placing each animal in both boxes for 20 min on the day before
training. Training on the subsequent day occurred in a single 18
min session during which the animals received three tone-shock
pairings after 4 min, 10 min, and 16 min. The CS presentation
consisted of a pulsed (0.7 sec on, 0.3 sec off) 20 sec tone.
Coincident with the offset of the tone, shock (US) was presented
for 500 msec. The following day the animals were placed in the
novel chamber where they received the same presentation of tones
but without the shock. (As it was critical to measure performance
during acquisition, tests of the tone in the novel chamber are not
reported here. This test was conducted simply to maintain compa-
rability to previous work in our laboratory.)
To quantify conditioned fear responses, the animal’s behavior
was videotaped and both the time spent freezing 20 sec prior to the
initiation of the tone as well as freezing during the tone was
determined. Freezing was defined as period of 1 sec or greater
when all of the animal’s paws remained anchored in-place to the
grid floor and the animal’s ears were retracted. Freezing responses
were scored by an observer who was unaware of the animal’s
performance on other behavioral tests. Conditioned responses to
the CS were defined as freezing during the tone presentation minus
freezing prior to the tone. For purpose of ranking the animals, CS
freezing during the second training trial was used.
4WASS ET AL.
AQ: 8
AQ: 9
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

Odor guided discrimination. Rodents rapidly learn to use
odors to guide appetitively reinforced behaviors. Here, mice
learned to navigate a square field in which unique odor-marked
(e.g., almond, lemon, mint) food cups were located in three cor-
ners. Although food was present in each cup, it was only accessible
to the animals in the cup marked by mint odor. An animal was
placed in the empty corner of the field, after which it explored the
field and eventually retrieved the single piece of available food. On
subsequent trials, the location of the food cups was changed, but
the accessible food was consistently marked by the same odor,
mint. On successive trials, animals required less time to retrieve
the food and made fewer search errors (see below) directed at food
cups in which food was not available. With this procedure, near
errorless performance is typically observed within 3– 4 training
trials.
A black Plexiglas 60 cm square field with 30 cm high walls was
located in a dimly lit (10 fc) testing room with a high ventilation
rate (3 min volume exchange). Three 4 42.0 cm (l, w, h)
aluminum food cups were placed in three corners of the field. A
food reinforcer (30 mg portions of chocolate flavored puffed rice)
was placed in a 1.6 cm deep, 1 cm diameter depression in the
center of each cup. The food in two of the cups was covered (1.0
cm below the surface of the cup) with a wire mesh so that it was
not accessible to the animal, while in the third cup (the “target”
cup), the food could be retrieved and consumed. A cotton-tipped
laboratory swab, located between the center and rear corner of
each cup, extended vertically 3 cm from the cups’ surface.
Immediately prior to each trial, fresh swabs were loaded with 25
l of either lemon, almond, or mint odorants (McCormick flavor
extracts). The mint odor was always associated with the target food
cup. It should be noted that, in pilot studies, the odor associated
with food was counterbalanced across animals, and no discernible
differences in performance were detected in response to the dif-
ferent odors. On the test day, animals received four training trials
in the field with three food cups present. On each trial, an animal
was placed in the empty corner of the field. On Trial 1, the
reinforcing food was available to the animal in the cup marked by
mint odor. The trial continued until the animal retrieved and
consumed the food from the target cup, after which the animal was
left in the chamber for an additional 20 sec and then returned to its
home cage to begin a 6 min ITI. On Trials 2– 4, the location of the
food cups was rearranged, but the baited cup was consistently
marked by the mint odor. Both the corner location of the mint odor
and its position relative to the remaining odors were changed on
each trial. On each trial, the errors committed by the mouse were
recorded. An error was recorded any time an animal made contact
with an incorrect cup, or its nose crossed a plane parallel to the
perimeter of an incorrect cup.
Fast mapping procedure. Two weeks after the completion
of the previous test, animals’ capacity for deductive reasoning was
determined based on their performance in a fast mapping task
(illustrated in Figure 1). On the first day of training, animals were
placed in a square open field constructed of white Plexiglas (46 
46 13 cm) for 15 min. Located within the field were eight
uniformly distributed objects (small plastic figurines, Mattel Corp.,
01-09TL) that would later be used to form paired associates. This
initial exposure was intended to familiarize animals to these ob-
jects, as two of the eight objects would otherwise only be encoun-
tered on the final (critical) test trial. (This procedure would be
repeated on the day prior to the critical test.)
On Day 2 of training, animals were acclimated to the training
apparatus which was constructed of black Plexiglas. The training
apparatus consisted of a start box measuring 20 14 20 cm and
a field portion. The walls (46 cm in length and 20 cm high) of the
field portion radiated from the start box at an approximate 145
degree angle and were connected to the back wall of the field
which was measured 60 cm in length and 20 cm high. Located
within the start box was a square block of black Plexiglas mea-
suring 7.5 7.5 1.5 cm which was fastened to the base of the
apparatus by Velcro. In the center of the block was a food port
measuring 1.5 cm in diameter and 1 cm in depth. This served as the
reinforcer (45 mg BioServ dustless precision pellets) location.
Three identical blocks were also placed in the field portion of the
apparatus. An inaccessible food port (to which a food reinforcer
was loaded in all cups) measuring 1 cm in diameter was drilled
into the side of each food cup. A 2 mm hole was drilled directly
into the center of the inaccessible food port until it protruded into
the accessible food port. Food placed in the port served as an odor
stimulus to ensure that the smell of food in the target cup could not
differentially guide the animal’s behavior during training or test-
ing. During this acclimation phase all of the possible food loca-
tions (a total of four locations) were baited. Initially the animals
were confined to the start box for a period of five minutes. After
this time period had elapsed the clear Plexiglas door (separating
the start box from the field) was removed allowing the animal to
venture into the field portion of the apparatus where it was con-
fined for another period of 5 min.
On Day 3, the training process was begun. The animals were
trained in two distinct phases. During the first phase, animals were
placed in the start location with a sample object (chosen from the
eight figures that the animals were exposed to on Day 1) that had
a reinforcer located in a recessed cup beneath it. Once the animal
retrieved the reinforcer, the door was opened and the animal was
allowed to venture into the field portion where only the sample
object’s paired associate (target object) was located. Under this
target object was a second reinforcer. Training using this same
procedure was continued (with a 6 min intertrial interval) for the
two additional paired associates. This procedure was repeated for
4 days (a total of eight exposures to each of the three stimulus
pairs), with the order of exposure to each pair randomly deter-
mined on each day. After paired associate training (Phase 1) was
complete, Phase 2 of training began on the subsequent day, at
which time animals were exposed to an object from one pair (the
sample) and were allowed to collect food from under its paired
associate (the target). However, on these trials the target object
was placed in a field containing two distracter objects (objects that
had previously been paired with a different sample object). Six
such training trials were conducted per day such that each pair of
objects was encountered two times each day. This phase of training
spanned 8 days (a total of 16 additional exposures to each stimulus
pair), at the end of which all animals performed near errorlessly,
choosing the correct target on at least five of the last six trials.
One day after the completion of Phase 2 paired-associate train-
ing, the animals were again placed in the open field (as described
above) with the six paired-associate objects and two novel objects
(previously encountered only in the open field) uniformly distrib-
uted throughout the field. Again, this was intended to moderate
5
REASONING AND GENERAL LEARNING
F1
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

any differential responding to the two novel objects that animals
would encounter on the critical fast mapping test trial. On the
subsequent day, animals received a single test trial to assess their
capacity for fast mapping. A fast mapping test trial consisted of
exposure to a novel (previously unencountered) sample object and
a novel target object placed within a field of three objects, two of
which were previous paired associates (i.e., which had acquired
specific meaning). After exposure to the novel sample and retrieval
of the reinforcer located beneath it, the animals were allowed to
explore the field of three objects (two previously trained on one
novel to the test context). Target choices were then recorded until
the piece of food located under the novel object was collected.
Results and Discussion
Here, we assessed 41 genetically heterogeneous CD-1 mice on
a battery of five learning tasks (i.e., Lashley III maze, passive
avoidance, odor guided discrimination, Morris water maze, and
associative fear conditioning) designed to tax different sensory/
motor, and information processing systems. These tasks placed
unique sensory, motor, motivational, and information processing
demands on the animals, thereby maximizing the likelihood that
the underlying influence across all tasks was domain-independent
(i.e., “general”).
The performance of individual animals across all learning tasks
was first analyzed with a principal component analysis. This is a
variable reduction procedure that uses an orthogonal transforma-
tion to convert a set of independent observations (potentially
correlated variables) into a set of uncorrelated variables (i.e.,
principal components). The goal of this analysis is to provide a
reasonable characterization of the complete data set by reducing a
correlation matrix to the fewest number of factors that can describe
the pattern of correlations. The principal factor is that factor which
can explain the largest amount of variance. Previous work has
indicated that as much as 28 – 48% of the variance in performance
across the five learning tasks reported here can be accounted for by
a single factor (see Kolata et al., 2008, for a complete description).
In the present study, a principle component factor analysis of
animals’ performance on the five learning tasks (see Table 1)
indicated that performance on all tasks were influenced by a single
source of variance. That factor (eigenvalue of 1.72) accounted for
29% of the variance in the performance across all of the learning
tasks. From that analysis, a general learning factor score was
Figure 1. An illustration of the procedure for the “fast mapping” task. During Phase 1, animals were taught
to associate pairs of objects (“paired associates). In Phase 2, animals were allowed to find the relevant paired
associate within a field that contained several objects, all of which had previously undergone paired associate
training. In Phase 3, a “fast mapping” test was administered. On this test trial, animals were exposed to a novel
sample object, and then allowed to explore the test field which contained a set of familiar objects (ones that had
an established meaning based on prior paired-associate training) and one novel object. (Note: For simplicity, the
directional arrows illustrated above all point to a correct target object located in the center position of the test
field. During actual trials, the location of the correct target was randomly determined).
6WASS ET AL.
T1
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

calculated for each of the animals. A factor score is analogous to
an average zscore of an animal’s performance on the five learning
tasks, with each score weighted according to the individual tasks
loading on the primary (general learning) factor. Thus, an animal’s
factor score is a quantification of that animal’s position in the
distribution of general learning abilities.
In subsequent analyses, animals would be compared based on
their aggregate performance across all learning tasks. This was
accomplished by separating animals into groups comprised of
high, intermediate, and low factor scores (based on the principal
components analysis presented above). That is, the factor scores
(of each individual) were ranked, and the top, middle, and bottom
thirds of these ranked scores were used to construct groups of
animals representing high, intermediate, and low general learning
abilities. Figure 2 presents the learning performance on each
learning task of animals characterized as having high or low
general learning abilities. As is evident from this figure, animals so
classified are clearly distinguishable, not only according to their
aggregate performance, but also on individual learning tasks.
After completion of the learning battery, animals began training
on a task intended to assess fast mapping. Animals were first
required to learn associations between three pairs of objects (small
plastic figurines) in order to obtain a food reward, and their
capacity for fast mapping was then evaluated. A fast mapping test
trial consisted of exposure to a novel (previously untrained) sam-
ple object and a novel (previously untrained) target object that was
located in a field of three objects (where the remaining two objects
were previously paired associates). Target choices were then re-
corded until the reinforcer located beneath the novel object was
collected. If animals were employing inferential exclusion, they
should make the inference that since the sample object located in
the start box was novel, they should direct their choice toward the
novel object in the test field. The number of errors (incorrect target
choices) that the animals made was recorded and compared to their
factor scores (their aggregate performance in the learning battery,
where lower scores higher aggregate learning performance)
which revealed a significant correlation, r(39) .44, p.01,
indicating that animals with higher aggregate learning abilities also
made fewer fast mapping errors. This correlation was reflected in
marked differences in performance across animals of high, inter-
mediate, and low general learning abilities (see Figure 3). A
comparison of the fast mapping errors of these three groups using
a one way ANOVA revealed a main effect of group (i.e., learning
ability), F(2, 38) 5.53, p.008. An LSD post hoc analysis
revealed a significant difference between the best and the worst
learners, p.002. No other comparison was significant, although
a trend toward a difference was observed when the intermediate
learners were compared to the animals of low learning abilities,
p.09.
It is important to note that across the entire sample of 41
animals, 22 animals made no fast mapping errors, and on average,
animals made only .4 errors. This number of errors is far below
that which would occur by chance (where one [or potentially more
if errors were repeated] errors would be expected in a random
search), 
2
(41, 2) 22.87, p.0001, indicating that rodents are
indeed capable of nominal deductive reasoning as represented by
performance on a fast-mapping task.
One might interpret the results of this experiment as the conse-
quence of the animals simply approaching the novel object in the
field rather than deducing that the novel target object was the
correct choice in response to the novel sample object. However, it
should be reiterated that prior to and again at the completion of
paired-associate training, animals were exposed equally to all
sample and test objects to mitigate this potential influence. Nev-
ertheless, the “novel” test object was designated as novel based on
its having undergone no prior paired associate training, and as
such, was less familiar to the animals than were the two distractor
objects (that had undergone previous paired associate training). A
more complete test of the possibility that the relative novelty of the
test object, as opposed to fast mapping, could account for the
results presented here is provided in Experiment 1B.
Experiment 1B
With a task that was procedurally and conceptually analogous to
fast mapping, in Experiment 1A, we ostensibly demonstrated that
mice were capable of making choices based on inferential exclu-
sion. Furthermore, performance on the fast-mapping task was
significantly correlated with animals’ aggregate performance in
battery of learning tests, suggesting that nominal reasoning and
general learning abilities were jointly regulated. However, on the
critical fast mapping test trial, a correct choice was indicated by an
animal approaching a test stimulus (in response to a novel sample)
that was less familiar than the comparison objects (that had un-
dergone paired-associate training). A correct choice was presumed
to reflect the animal’s decision that in a field of familiar objects,
the novel sample must be associated with the more novel test
object. In that task, animals had been preexposed (in a nontest
environment) to all stimuli so as to reduce the nominal novelty of
the sample and test stimuli at the time of test. Nevertheless, at the
time of the critical test, the novel test stimulus was less familiar to
the animals than were the previously trained stimuli, raising the
possibility that animals were simply attracted to the more novel
stimulus, rather than selecting that object through the process of
inferential exclusion. Given that the animals were food deprived
and the familiar objects in the test field were previously paired
with food, this seemed an unlikely explanation for the animals’
performance. Nevertheless, based on the procedure in Experiment
1A, this possibility cannot be excluded.
In order to ascertain whether the subjects’ performance on the
fast mapping test in Experiment 1A was influenced by a propensity
for novelty seeking, in Experiment 1B we implemented a relevant
control procedure into the fast mapping task. Using a group of
Table 1
Factor Loadings From The Principal Components Analysis
(n 41) For Animals’ Performance On The Five Learning
Tasks In Experiment 1A
General learning factor
Lashley III Maze .60
Fear Conditioning .52
Passive Avoidance .72
Odor Discrimination .61
Morris Water Maze .40
eigenvalue 1.72
% variance .29
7
REASONING AND GENERAL LEARNING
F2
F3
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

experimentally naı¨ve animals, this experiment followed the gen-
eral procedure of Experiment 1. However, at the time of testing,
our sample of mice was divided into two groups (with statistically
equal general learning abilities). One group received a standard
fast-mapping test trial, where a novel sample stimulus was fol-
lowed by the subjects’ choice among a field of two familiar objects
(previously paired associates) and one novel object. On this test
trial, a second group was exposed to a familiar sample, two
D: Fear Conditioning
Trials
321
Freezing During CS (Sec)
4
6
8
10
12
High Aggregate Learning Abilities
Low Aggregate Learning Abilities
B: Odor-Guided Discrimination
Trials
1234
Errors to Locate Reinforcer
0
2
4
6
8
10
12
14
16
18
High Aggregate Learning Abilities
Low Aggregate Learning Abilities
E: Passive Avoidance
Aggregate Learning Abilities
LOW HIGH
Post/Pre-Step Down Latency
1
2
3
4
5
C: Morris Water Maze
Trials
12345678910
Latency (sec) to Locate Platform
10
20
30
40
50
60
70
80
High Aggregate Learning Abilities
Low Aggregate Learning Abilities
A: Lashley III Maze
Trials
12345
Errors to Locate Reinforcer
5
10
15
20
25
30
High Aggregate Learning Abilities
Low Aggregate Learning Abilities
Figure 2. Performance on individual tasks of animals of highest and lowest general cognitive abilities. General
learning abilities were determined by factor scores (of individual animals) derived from a principal component
analysis of the acquisition data from all five learning tasks. Illustrated is the mean performance of animals of
high and low general learning abilities (for clarity, animals of intermediate abilities are not illustrated.) Animals
with high general learning abilities outperformed animals of low general learning abilities in each of the five
individual tasks (Lashley Maze [A], odor-guided discrimination [B], Morris Water Maze [C], fear conditioning
[D], and passive avoidance (E]). Brackets indicate standard error of the mean.
8WASS ET AL.
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

familiar choice objects (one of which was the correct paired
associate), and one novel object. If the choice of the novel object
was directed by the animals’ propensity to choose novelty, both of
the groups in this experiment should choose the novel object on
this test trial. However, if animals’ performance was guided by
inferential exclusion, only animals administered the standard fast-
mapping test should choose the novel object. In addition to an
assessment of novelty-seeking as an explanation of animals’ fast
mapping performance, Experiment 1B also served to assess the
reliability of the results described in Experiment 1A.
Method
Subjects. A new sample of 25, experimentally naı¨ve geneti-
cally heterogeneous male CD-1 outbred mice were obtained from
Harlan Laboratories (Indianapolis, IN) at 45 days of age. Housing
and maintenance conditions were identical to Experiment 1A.
Learning battery. Animals were first assessed in the learning
battery as described in Experiment 1A. Testing in the learning
battery provided a second opportunity to assess the relationship
between fast mapping performance and general learning abilities.
Furthermore, having characterized each animal’s aggregate learn-
ing performance, we were able to assign an equal distribution of
learning abilities to both of the groups represented in this experi-
ment.
Fast mapping. Two weeks after the completion of the learn-
ing battery, the initial group of 25 animals was separated into two
groups which will be referred to as “Group FM” (Fast Mapping)
and “Group NS” (Novelty Seeking). Group FM (n13) animals
would undergo the fast mapping training and testing as described
in Experiment 1A. Group NS (n12) would undergo a near
identical training and testing procedure with one critical difference
intended to ascertain whether Group FM’s performance was due to
their propensity for novelty seeking or if these animals’ perfor-
mance was indicative of fast mapping. Specifically, on the critical
test trial Group NS would be exposed to a familiar sample,
followed by a choice from among three test objects that included
the correct paired associate, an incorrect paired associate, and a
novel object.
On the first day of training, all 25 animals were placed in a
square open field constructed of white Plexiglas (46 46 13
cm) for 15 min. For Group FM, there were eight uniformly
distributed objects (small plastic figurines, Mattel Corp., 01-09TL)
located within the field, of which six would later be used to form
paired associates. The remaining two objects would later be used
as the critical (novel) sample and test objects. In contrast, the test
objects were absent for Group NS so that the remaining objects in
the field would be the six objects that would later form paired
associates. This initial exposure was intended to familiarize both
the experimental and control groups to the objects that would later
constitute paired associates. By not exposing Group NS to the two
test objects, it insured that those objects would be entirely novel at
the time of testing (so as to maximize any influence of novelty
seeking).
On Day 2 of training, all 25 animals were acclimated to the
training apparatus (which was identical to the one used in Exper-
iment 1A). Animals then underwent paired-associate training as in
Experiment 1A, and on the day following completion of this
training, the animals were again placed in the open field (as
described above). Group NS was only exposed to the six paired-
associate objects and Group FM was exposed to those six objects
as well as the two that would comprise the sample and test stimuli
on the critical test trial.
On the subsequent day, Group FM received a single test trial to
assess their capacity for fast mapping. The fast mapping test trial
consisted of exposure to a novel (previously unencountered in
paired-associate training) sample object and a novel target object
placed within a field of three objects, two of which were previous
paired associates. After exposure to the novel sample and retrieval
of the reinforcer located beneath it, the animals were allowed to
explore the field of three objects (two previously trained and one
novel to the test context). Target choices were then recorded until
the piece of food located under the novel object was collected. For
Group NS, a previously reinforced sample object was placed in the
start box and its paired associate was located within the field with
two distracter objects (a total of three objects). One of the dis-
tracter objects was an incorrect paired associate whereas the sec-
ond object was a novel object to which the animal had not
previously been exposed. Target choices were recorded until the
animal chose the correct paired associate.
Results and Discussion
A principal component analysis of the 25 animals’ performance
on the five learning tasks indicated that performance on all of the
tasks was influenced by a single source of variance (see Table 2).
That factor accounted for 32% (eigenvalue 1.9) of the variance in
the animals’ performance across all learning tasks. From that
factor analysis we were able to derive factor scores (as described
above) for each individual animal. Those factor scores (which
ranged from:
Average Errors
0.2
0.4
0.6
0.8
1.0
High Intermediate Low
p < .01
GENERAL LEARNING ABILITY
Figure 3. Fast mapping test performance, Experiment 1A. Three groups
of animals were formed based on the upper, middle, and bottom third of
factor scores (reflective of general learning performance) obtained from the
principal component analysis of learning test performance in Experiment
1A. Plotted is average number of errors (standard error) on the fast
mapping test trial of the animals that performed best (High), intermediate,
and worst (Low) on the battery of learning tasks. For this task, one error
(on average) would be expected in a random search (assuming that re-
peated errors were not committed, in which case, the number of errors
could increase).
9
REASONING AND GENERAL LEARNING
T2
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

2.39 to 1.83, where lower values indicate better aggregate
learning performance) were then used to divide the subjects into
two groups (Groups FM and NS) of roughly equal general learning
abilities (with an equal representation of animals of high, interme-
diate, and low general learning abilities). Following their assign-
ment to groups, the mean factor score of these two groups was .034
and .048 (Groups FM and SM, respectively) and the groups did not
differ statistically, t(23) .03.
If animals’ propensity for novelty seeking determined their
performance on the putative fast mapping test trial, then Group NS
would be expected to choose the novel object in the test field
despite the presence of a familiar sample and test stimulus (the
correct paired associate) on the critical test trial. However, 10 out
of the 12 animals in Group NS made no errors, choosing the
correct paired associate test object on their first choice. Of the two
subjects that did make an error, they incorrectly chose the dis-
tracter paired associate object, not the novel test object, and then
proceeded to the correct paired associate. Of the 12 animals, 0/12
chose the novel test stimulus on either their first or second choice.
This number of choices of the novel stimulus is below that which
would occur through even a random search, 
2
(2) 14, p.001,
indicating that the animals are utilizing a process that was inde-
pendent of novelty seeking to guide their behavior.
Unlike Group NS, Group FM was presented with both a novel
sample object and a novel test object among a field containing two
familiar objects (with a history of paired associate training). In-
ference by exclusion would dictate that under these circumstances,
animals would be disposed to choose the novel test stimulus. In
this instance, six of 13 animals chose the novel test stimulus on
their first choice, but given the relatively small sample size, this
pattern was not significantly different than chance. However, an
inspection of the subgroup of these animals with the highest
general learning scores (n8) revealed that these animals per-
formed well above chance levels, choosing the correct (novel) test
object on six out of eight first choices (from the set of three
possible choices), 
2
(2) 6.24, p.05. This pattern, like that
described in Experiment 1, indicates a capacity of these animals
for inference by exclusion. In addition, this result indicates the
importance of considering the performance of individual animals
when assessing animals’ capacity to perform on a task that is at the
upper limits of their cognitive abilities.
Having demonstrated that animals’ fast mapping performance
was not related to their propensity for novelty seeking, a subse-
quent analysis of Group FM’s performance (errors) in the fast
mapping task was compared to their aggregate performance across
all learning tasks (i.e., their factor scores). That comparison re-
vealed a significant correlation, r(11) .67, p.02, which
indicates that animals with higher general learning abilities made
fewer fast mapping errors. A further analysis examining the top
third of the distribution (animals with high GLA scores), the
middle third (intermediate GLA scores) and bottom third (animals
with low GLA scores) of the distribution using a one way ANOVA
revealed a main effect of group F(2, 9) 5.39, p.03. An LSD
post hoc analysis revealed a significant difference between animals
of high general learning abilities and low general learning abilities,
p.01. No other comparisons were significant, but a trend toward
a significance was observed when animals of intermediate learning
abilities and low learning abilities were compared, p.056. These
results are illustrated in Figure 4.
Experiment 2
It was established in Experiment 1A and 1B that the general
learning ability of mice was correlated with their performance on
a nominal deductive reasoning (fast mapping) task. The aim of
Experiment 2 was to assess whether animals’ general learning and
inductive reasoning abilities were also coregulated. For this pur-
pose, animals that had been characterized for general learning
performance in Experiment 1A were used here (so as to compare
these abilities within a single group of animals). After the com-
pletion of Experiment 1A, the animals that participated in that
experiment were assessed on a task which could be most effi-
ciently performed through the application of inductive reasoning.
To assess inductive reasoning abilities, we developed a task in
GENERAL LEARNING ABILITY
High Intermediate Low
Average Errors
0.5
1.0
1.5
2.0 p < .01
Figure 4. Fast mapping test performance, Experiment 1B. Three groups
of animals were formed based on the upper, middle, and bottom third of
factor scores (reflective of general learning performance) obtained from the
principal component analysis of learning test performance in Experiment
1B. Plotted is average number of errors (standard error) on the fast
mapping test trial of the animals that performed best (High), intermediate,
and worst (Low) on the battery of learning tasks. For this task, one error
(on average) would be expected in a random search (assuming that re-
peated errors were not committed, in which case, the number of errors
could increase).
Table 2
Factor Loadings From The Principal Components Analysis
(n 25) For Animals’ Performance On The Five Learning
Tasks In Experiment 1B
General learning factor
Lashley III Maze .62
Fear Conditioning .76
Passive Avoidance .51
Odor Discrimination .32
Morris Water Maze .55
eigenvalue 1.9
% variance .32
10 WASS ET AL.
F4
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

which animals could discern an efficient search strategy based on
their experience with the overall structure of a “decision-tree” (or
binary) maze (graphically represented in Figure 5). Binary trees
are used in operations research to identify strategies that are most
efficient in reaching a goal. (A colloquial version of the decision
tree is exemplified by the process utilized to identify an object in
the game of “20 Questions.”) Of note, this task is not simply a
maze-learning task, such as the Lashley III Maze used in the
learning battery. In a typical egocentric maze (like the Lashley
Maze), the food is consistently located in a goal box and the
animals learn that one path is most efficient in reaching that goal.
Once learned, execution of a route does not involve active search-
ing. By comparison, in this binary tree maze, both the location and
amount of food is randomly distributed throughout the maze at
various choice points, and there are many possible routes (of
varying degrees of efficiency) with which to explore the maze to
find the available food. Thus, by design, the requirements of this
maze promotes the implementation of a search strategy. It is the
efficacy of that strategy that will serve as our index of inductive
reasoning.
Method
Subjects. The 47 animals that served in Experiment 1A were
used here. Housing and maintenance conditions were as described
above.
Decision tree maze. Two weeks following the completion of
Experiment 1A, animals began training in the decision tree. As
depicted in Figure 5, the maze walls, floor, and doors were
constructed of black Plexiglas, and a transparent sheet of Plexiglas
covered the top of the maze. The maze consisted of a start box and
a series of bifurcating arms. The maze bifurcated along each
branch at seven symmetrically arranged locations and terminated
in dead ends (“leaves”). Other than at the first bifurcation (which
divided the maze into two symmetric halves), each bifurcation and
each leaf constituted a “node” at which a reinforcer (20 mg
BioServe chocolate flavored pellet) could be hidden in a recessed
cup in the floor.
The apparatus consisted of a start box (8 67 cm) located
at the base of the decision tree which was separated from the maze
by a removable door (represented by a dashed line in Figure 5).
Extending from the start box was the first alley of the decision tree
which, like all other alleys, measured 22 67 cm. At the end
of the first alley was the first bifurcation point in which there was
no food port located. The locations of the food cups are depicted
in Figure 5 as dots located at each bifurcation point (except the
initial bifurcation point) as well as at the end of each leaf. The food
cups were recessed holes in the base of the maze. If baited with a
reinforce, the reinforcer would be below the level of the floor
ensuring that the animal was unable to see the reinforcer as it
approached the port. Each food port measure 1 cm in diameter and
1.5 cm in depth.A2mmhole was drilled directly into the center
of each cup which protruded through the base of the maze. This
hole served as an odor port for inaccessible food reinforcers (80
mg) that were placed beneath each food cup (to provide a uniform
distribution of olfactory cues).
Forty-eight hours prior to the acclimation day the animals were
food deprived and allowed 90 min of free access to food toward
the end of their light cycle. Also, on the day prior to acclimation,
the animals were given three reinforcers to familiarize them
with the novel reinforcers to be used in the binary maze. On the
acclimation day the animals were confined to the start box for a
period of 30 sec. After 30 sec, the removable door separating the
start box from the first alley was removed, allowing the animal to
enter the maze. During this first exposure to the maze, all 14 nodes
were baited. The animal was allowed 20 min to freely navigate the
maze and find all possible reinforcers before being brought back to
the vivarium.
On 10 subsequent testing days (i.e., trials), as few as four and as
many as eight nodes were randomly baited, with a minimum of
two baited nodes on each half of the maze (all animals received the
same baited nodes). Since the actual location and number of baited
nodes varied randomly across trials, it was advantageous to the
animals to adopt a strategy for inspecting all nodes with equal
likelihood. On these critical test days, the animal was once again
confined to the start box for 30 seconds, released, and allowed to
explore the maze until all the reinforcers were retrieved and all
nodes were crossed at least once.
To determine if the animals had implemented a flexible search
strategy or were simply following a rote path, after the 10th day of
testing, a new procedure was instituted. Three trials were admin-
istered (on successive days) where when an animal reached its first
terminal leaf, the second level junction adjacent to the one occu-
pied by the animal was blocked. There were four possible locations
that could have potentially been blocked and these are represented
as dashed lines in Figure 5. As above, animals were allowed to
continue through the maze until all food had been retrieved.
Figure 5. Illustration of the binary decision maze. Here, the animals’ task
was to navigate the maze so as to inspect every potential node (labeled
1–14) for a payoff (a piece of food recessed in the floor). At the beginning
of each trial, food was randomly placed in 4 –8 locations. Since the animal
could not know the location of food or the number of goal locations that
were actually baited, an efficient search would inspect all nodes with
minimal duplication of effort. Using an optimal search strategy, the animal
would pass a maximum of 24 nodes (as would be required were the animal
to search, without unnecessary duplication, every node in one half of the
maze, exit that side of the maze, then search every node in the other half
of the maze).
11
REASONING AND GENERAL LEARNING
F5
AQ: 10
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

To assess animal’s search efficiency, “streak” lengths were
recorded (i.e., the number of nodes crossed prior to an animal
unnecessarily crossing a node that had already been visited) for
both stages of testing (Trials 1–10 and 11–13). For the first stage
of testing (Trials 1–10), a maximally efficient search would require
24 node crossings. During the second stage (Trials 11–13), an
efficient search would result in 18 nodes being crossed (since one
branch was rendered inaccessible to the animals).
Results and Discussion
In their initial 1– 4 exposures to the decision tree maze, no
systematic pattern of exploration could be detected across the
group of animals or within individuals (i.e., the animals’ pattern of
behavior suggested a disorganized random search). However,
within 4 – 6 trials, the patterns of individual animals stabilized and
remained stable for the remaining 4 days of testing (see Figure 6).
Although several animals performed at optimal efficiency during
the last four trials, other animals’ performance remained unsys-
tematic. Streak lengths ranged from 4 –24 (maximal efficiency) on
each of the last four trials (indicative of wide variability in ani-
mals’ performance).
The average streak length on the last four trials was compared to
factor scores (indicative of general learning abilities) obtained
from the principal component analysis of learning performance
(described in Experiment 1A). These two independent measures
were significantly correlated, r(45) .46, p.01 (Figure 7A),
indicating that the more efficient search (as indicated by longer
streaks) was associated with better aggregate learning abilities (as
indicated by lower factor scores). As evident in Figure 7B, the
streak lengths differed between animals of high, intermediate, and
low general learning abilities (the bottom, middle, and top third of
factor scores), F(2, 44) 7.65, p.001. A LSD post hoc analysis
revealed a significant difference between the streak lengths of
Trials
12345678910
Average Streak
6
8
10
12
Figure 6. Performance in the binary decision tree. Plotted is the average
streak length of all 47 animals tested in this maze, where a streak of 24
would reflect optimal efficiency. Animals’ performance was initially er-
ratic, but stabilized within six trials and remained stable thereafter. Al-
though several animals performed at optimal efficiency during the last four
trials, other animals exhibited unsystematic performance. Streak lengths
(across animals) ranged from 4 –24 on each of the last four trials. Brackets
indicate standard error of the mean.
Average Streak (Three Probe Trials)
4 6 8 10 12 14
Factor Score
-2
-1
0
1
2
Average Streak (Last Four Training Trials)
4 6 8 10121416182022
Factor Score
-2
-1
0
1
2
r= -.46
r= -.56
High Intermediate Low
Average Streak
2
4
6
8
10
12
A
B
C
Figure 7. Individual animals’ performance in the binary decision maze is
predicted by their aggregate (general) learning ability.Panel A: Factor
scores for each animal were derived from a principal component analysis
of all animals’ performance on five learning tasks. These scores reflect
animals’ aggregate performance across all five tasks. (Note that lower
factor scores better aggregate learning performance.) A significant
correlation was observed between animals’ factor scores and the number of
node crossings prior to unnecessarily crossing a node (“streak” perfor-
mance) on the last four (of 10) test trials. Thus, the efficacy of an animal’s
search (a form of inductive reasoning) was predicted by their general
learning ability. Panel B: Three groups of animals were formed based on
the upper, middle, and bottom third of factor scores (reflective of general
learning performance). The average streak length (indicative of search
efficacy) differed across these three groups. Plotted is the animals that
performed best (High), intermediate, and worst (Low) on the battery of
learning tasks. Brackets indicate standard errors. Panel C: After an animal
reached its first low-level terminal node, the adjacent entry point was
blocked with a sliding door. This was intended to disrupt any rote path (i.e.,
algorithmic strategy) that an animal may have developed in lieu of com-
prehension of the overall structure of the maze. Plotted is the average streak
during three such probe trials against factor scores obtained from the
principal component analysis of learning performance. (Note that lower
factor scores better aggregate learning performance.) Again, a signifi-
cant correlation between general learning abilities and search efficacy was
observed.
12 WASS ET AL.
F6
F7
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

animals of low general learning abilities and animals of either
intermediate or high abilities, p.01. No difference in streak
length was observed between animals of high and intermediate
learning abilities.
Following the initial 10 trials, we determined if animals were
relying on rote paths through the maze or whether they were
engaging in an active search of the maze (a requisite for inductive
reasoning). To make this determination, each animal was allowed
to begin its exploration of the maze, and upon making its first entry
into a second level branch, the adjacent branch was blocked by
lowering a black guillotine door. Had an animal been following a
rote (but nominally efficient) path through the maze, this manip-
ulation would have disrupted the utilization of that rote path.
Animals’ streak lengths under these conditions were assessed on
three trials. Despite the imposed disruption of the path, the corre-
lation between the animals’ average streak and factor scores (ag-
gregate learning performance) was still strong, r(45) .56, p
.01 (Figure 7C). The average streak lengths for these three trials
differed between animals of high, intermediate, and low general
learning abilities, F(2, 44) 9.44, p.001 (means sem 
10.2 0.86, 7.9 0.68, and 6.0 .42, respectively, where a
perfect streak 18). LSD post hoc analysis revealed a significant
difference between animals of high and intermediate abilities, p
.02, as well as between animals of high and low abilities, p.001.
Lastly, a significant difference between animals of intermediate
and low abilities was observed, p.05.
The degree to which an animal can devise an efficient strategy
to search a maze in which the location and number of reinforcers
continuously changes is at least nominally indicative of inductive
reasoning. While numerous search strategies may be utilized to
navigate this maze, one search strategy in particular, the depth-first
search, would be the most efficient. Other search strategies such as
the breadth-first search or random searching would be relatively
inefficient strategies owing to their necessitating unnecessary node
crossings. It should be noted that performance in this task should
not be considered to be exclusively an expression of a learned
route, as of the various number of routes that could be learned, no
single route is most efficacious, and furthermore, would be se-
verely disrupted by blocking the animals’ path. Also, throughout
this task the number of reinforcers available as well as the loca-
tions of the reinforcers varied each day, ensuring that the animal
wasn’t learning fixed reinforcer locations. These manipulations,
therefore, tested an animal’s ability to implement an efficient
search strategy in an environment that was explicitly unstable.
We have suggested that a more intelligent animal would be able
to form a more complete representation of the structure and re-
quirements of the maze, and would thereby implement a more
efficient search strategy. It should also be noted that other factors
such as the efficacy of an animal’s working memory may have
influenced their performance in this task. Working memory spe-
cifically has been shown to correlate with general learning abilities
in mice as well as in humans (Buehner, Krumm, & Pick, 2005;
Engle et al., 1995; Kolata et al., 2005, 2008; Light et al., 2010;
Sub, Oberauer, Wittman, Wilhelm, & Schulze, 2002). However,
the implementation of an efficient search strategy in this task
would minimize any reliance on working memory. It is also
recognized that performance in this maze would be described in
other contexts as a form of “foraging” (for review, see Rashotte,
O’Connel, & Djuric, 1985). Thus, an animal’s reasoning abilities
may directly relate to the animal’s performance in a foraging task
and have obvious and direct implications for survival in more
ethologically relevant environments.
Upon completion of Experiment 2, an additional principal com-
ponent analysis was conducted that included the learning data
reported in Experiment 1A, fast mapping performance from Ex-
periment 1A, and decision tree performance from Experiment 2.
(Data from Experiment 1B was not included in this analysis as it
was obtained from a separate sample of animals that were not
assessed in the decision tree maze.) Only animals (n41) that
contributed to both the fast mapping task (Experiment 1A) and the
decision tree task (Experiment 2) were included in this analysis. A
single factor accounted for 27% (eigan value 1.92) of the
variance across all performance measures, and performance on
both the decision tree and fast mapping tasks loaded moderately
and in the same direction as performance on all of the learning
tasks (see Table 3). This analysis suggests that a single underlying
source of variance influences performance on all of these diverse
cognitive tasks.
General Discussion
Over a century ago, Spearman reported the existence of a
general intelligence factor in humans (Spearman, 1904). The con-
cept of general intelligence has facilitated studies of individual
differences in the expression of this ubiquitous cognitive trait
(Jensen, 1998). Recent work with nonhuman animals has impli-
cated the existence of a general learning ability in genetically
heterogeneous mice, as well as individual variations in this trait
(Galsworthy et al., 2002; Kolata, Light, Grossman, Hale, & Mat-
zel, 2007; Kolata et al., 2005; Locurto et al., 2003; Matzel et al.,
2006). Although it has been asserted that this general learning
ability mice is psychometrically and structurally analogous to
“intelligence” in humans (Kolata et al., 2008; see Blinkhorn, 2003,
for commentary), it was not previously known whether general
learning abilities in mice were coregulated with the animals’
capacity for reasoning. Since reasoning is considered a central
function of intelligent behavior, here we aimed to address this
issue, using tasks based on ones that are often asserted to reflect
the capacity of humans for reasoning.
Human reasoning abilities have been found to be highly predic-
tive of a person’s general intelligence. It is in this regard the
Table 3
Factor Loadings From The Principal Components Analysis
(n 41) For Performance On The Five Learning Tasks As Well
As Fast Mapping Performance (Experiment 1A) and The Average
Streak Length In The Decision Tree Maze (Experiment 2)
General cognitive factor
Lashley III Maze .23
Fear Conditioning .55
Passive Avoidance .28
Odor Discrimination .8
Morris Water Maze .68
Decision Tree .49
Fast Mapping .31
eigenvalue 1.92
% variance .27
13
REASONING AND GENERAL LEARNING
T3
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

Raven’s Progressive Matrix (RPM) test has been asserted to be one
of the purest measures of general intelligence (Babcock, 1994; Fry
& Hale, 1996; Dawson, Soulieres, Gernsbacher, & Mottron, 2007;
Jensen, 1998), and has largely supplanted many historically prev-
alent adult intelligence tests. The RPM measures one’s ability to
infer rules, think rationally, reason by analogy, and to organize
spatial information into related wholes. Therefore, if the purest
measure of human general intelligence was based on one’s ability
to reason, and nonhuman animals’ general learning abilities were
in fact indicative of intelligence, it would follow that the reasoning
and general learning abilities of mice should be positively corre-
lated.
Following Aristotle, it is often asserted that reasoning can take
one of two general forms. In the first, one attempts to understand
the “whole” by considering only the component parts. In the
second, one attempts to characterize a class of objects by consid-
ering the common features of each object in that set. To assess
reasoning in laboratory mice, we devised two novel tasks which
reflect each of these forms of rational behavior. First, animals’
performance was assessed on a fast mapping task. Fast mapping is
a mental process whereby a new concept can be learned based on
a logical inference derived from a single exposure to incomplete
information. This corresponds with what is described as “deduc-
tive” reasoning, that is, an attempt to characterize a class of objects
by considering the common features of each object in that set. Fast
mapping is believed to play a critical role in the extraordinarily
rapid and seemingly effortless acquisition of information during
early human development, and explains (in part) the prodigious
rate at which children gain vocabulary (Carey & Bartlett, 1978). In
addition to fast mapping, animals’ performance was assessed in a
“decision” or binary-tree maze. Decision trees are commonly used
in operations research, specifically in decision analysis, to identify
strategies that are most efficient in reaching a goal. While many
search strategies (or paths) could be utilized to visit every node in
the decision tree, the vast majority of these paths would lead to an
inefficient search, that is, one which unnecessarily retraces paths
or crosses goals that had already been explored. Thus, the degree
to which an animal can comprehend the whole structure of the
maze and implement that information from its current location
would be a reflection of a type of “inductive” reasoning.
Given the focus of the present work on what is described as the
capacity for “reasoning” by mice, it should be acknowledged that
more elemental psychological processes (e.g., associative learning)
have been proposed to underlie other apparent instances of rea-
soning in nonhuman animals (Dwyer, Starns, & Honey, 2009;
Haselgrove, 2010; see Shettleworth, 2010 for relevant discussion;
but see Lazareva & Wasserman, 2006 for response). Similar ex-
planations (devoid of reference to reasoning) could be applied to
many instances of presumed reasoning by humans. We concede
that what might nominally be described as “reasoning” may, at
least in some instances, actually be the product of more elemental
processes. Were this the case, the relationship between human
learning abilities and intelligence (as indexed by reasoning-based
psychometric tests) may reflect the correlation of two traits that are
dependent on the same (or related) underlying processes (e.g.,
associative learning). The same may be true in the present case,
where we have concluded that animals’ performance (on the fast
mapping and decision tree tasks) is dependent, at least in part, on
forms of reasoning. In this regard, it was not our intention to argue
that performance on any particular task is exclusively indicative of
the implementation of reasoning. Rather, the rationale for the
present experiments was that if performance by humans on a task
(e.g., fast mapping) is said to rely on reasoning, than the same
description can be applied to the analogous performance of a
nonhuman animal. Our goal was not to unequivocally demonstrate
the capacity for reasoning in laboratory mice, but rather, to deter-
mine if performance on tasks that are attributed to reasoning (as
they would be when performed by humans) are predicted by the
general learning abilities of these animals.
The findings of the current study indicate that nonhuman ani-
mals are indeed capable of behaviors that are at least nominally
indicative of various forms of reasoning (also see Blaisdell, Sawa,
Leising, & Waldmann, 2006; Dusek & Eichenbaum, 1997; Pov-
inelli, 2000; Pilley & Reid, 2011). More relevant to our present
purpose, the efficacy with which mice performed on the two
reasoning tasks reported here was directly predicted by their ag-
gregate performance across a battery of five diverse learning tasks.
These observations support the hypothesis that reasoning and
general learning abilities of mice are the mutual expression of a
core cognitive ability that among humans would be described as
“intelligence.”
The observation here that reasoning and general learning abili-
ties are correlated should be considered in light of previous ob-
servations that the general learning abilities of mice are predicted
by variations in selective attention and working memory capacity
(Kolata et al., 2005, 2008; for review, see Matzel & Kolata, 2010;
for analogous results from tests of humans, see Halford, Cowan, &
Andrews, 2007; Unsworth & Engle, 2007). To return to one of the
definitions of intelligence provided above, concepts of “intelli-
gence” are attempts to classify “the ability to understand complex
ideas, to adapt effectively to the environment, to learn from expe-
rience, to engage in various forms of reasoning, to overcome
obstacles by taking thought.” While this and similar definitions
were conceived to describe a human cognitive trait, this same
definition appears relevant in summarizing the performance of
mice on this diverse set of cognitive tasks. Thus, like humans, mice
appear to express individual variations in intelligence, and these
variations have profound functional consequences for the animals’
negotiation of their environments. Overall, our observation of the
coregulation of diverse cognitive abilities in mice suggests that
the operations of intelligence may have been evolutionarily
conserved across distant mammalian species (see also Banerjee
et al., 2010).
It is worth noting that subregions of the prefrontal cortex may
mediate “intelligence” through their regulation of attentional con-
trol and/or working memory capacity (Durstewitz, Seaman, &
Sejnowski, 2000; Kolata et al., 2010; Sawaguchi & Goldman-
Rakic, 1991; Thurley, Senn, & Luscher, 2008; for review, see
Matzel & Kolata, 2010). In fact, Kolata et al. (2010) have reported
that a cluster of dopamine D1-related genes in the prefrontal cortex
are overexpressed in animals of high versus low general learning
abilities. Since general learning abilities, attentional control, and
reasoning abilities appear to be coregulated, it is tempting to
speculate that dopaminergic signaling in the prefrontal cortex
might serve as one of the (among potentially other) determinants
of variations in intelligence.
14 WASS ET AL.
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

References
Babcock, R. L. (1994). Analysis of adult age differences on the raven’s
advanced progressive matrices test. Psychology and Aging, 9, 303–314.
doi:10.1037/0882-7974.9.2.303
Banerjee, K., Chabris, C. F., Johnson, V. E., Lee, J. J., Tsao, F., & Hauser,
M. D. (2010). General intelligence in another primate: Individual dif-
ferences across cognitive task performance in a New World monkey
(Saguinus oedipus). PLoS ONE, 4, e5883–5894. doi:10.1371/
journal.pone.0005883
Blaisdell, A. P., Sawa, K., Leising, K. J., & Waldmann, M. R. (2006).
Causal reasoning in rats. Science, 311, 1020 –1022. doi:10.1126/
science.1121872
Blinkhorn, S. (2003). Of mice and mentality. Nature, 424, 1004 –1005.
doi:10.1038/4241004a
Boysen, S. T., Berntson, G. G., Shreyer, T. A., & Quigley, K. S. (1993).
Processing of ordinality and transitivity by chimpanzees. Journal of
Comparative Psychology, 107, 208 –215. doi:10.1037/0735-7036
.107.2.208
Buehner, M., Krumm, S., & Pick, M. (2005). Reasoning working
memory attention. Intelligence, 33, 251–272. doi:10.1016/
j.intell.2005.01.002
Carey, S., & Bartlett, E. (1978). Acquiring a single new word. Proceedings
of the Stanford Child Language Conference, 15, 17–29.
Conway, A. R., Kane, M. J., & Engle, R. W. (2003). Working memory
capacity and its relation to general intelligence. Trends in Cognitive
Sciences, 7, 522–547. doi:10.1016/j.tics.2003.10.005
Dawson, M., Soulieres, I., Gernsbacher, M. A., & Mottron, L. (2007). The
level and nature of autistic intelligence. Psychological Science, 18,
657– 662. doi:10.1111/j.1467-9280.2007.01954.x
Durstewitz, D., Seamans J. K., & Sejnowski, T. J. (2000). Dopamine-
mediated stabilization of delay-period activity in a network model of
prefrontal cortex. Journal of Neurophysiology, 83, 1733–1750.
Dusek, J. A., & Eichenbaum, H. (1997). The hippocampus and memory for
orderly stimulus relations. The National Academy of Sciences, 94, 7109 –
7114. doi:10.1073/pnas.94.13.7109
Dwyer, D. M., Starns, J., & Honey, R. C. (2009). “Causal reasoning” in
rats: A reappraisal. Journal of Experimental Psychology: Animal Behav-
ior Processes, 35, 578 –586. doi:10.1037/a0015007
Engle, R. W., Laughlin, J. E., Tuholski, S. W., & Conway, A. R. A. (1999).
Working memory, short-term memory, and general fluid intelligence: A
latent variable approach. Journal of Experimental Psychology, 128,
309 –331. doi:10.1037/0096-3445.128.3.309
Fry, A. F., & Hale, S. (1996). Processing speed, working memory, and
fluid intelligence: Evidence for a developmental cascade. Psychological
Science, 7, 237–241. doi:10.1111/j.1467-9280.1996.tb00366.x
Galsworthy, M. J., Paya-Cano, J. L., Monleo´ n, S., & Plomin, R. (2002).
Evidence for general cognitive ability (g) in heterogeneous stock mice
and an analysis of potential confounds. Genes, Brain, & Behavior, 11,
88 –95. doi:10.1034/j.1601-183X.2002.10204.x
Gathercole, S. E., & Baddeley, A. D. (1989). Evaluation of the role of
phonological STM in the development of vocabulary in children, a
longitudinal study. Journal of Memory and Language, 28, 200 –213.
doi:10.1016/0749-596X(89)90044-2
Gettinger, M. (1984). Individual differences in time needed for learning: A
review of literature. Educational Psychologist, 19, 15–29. doi:10.1080/
00461528409529278
Halford, G. S., Cowan, N., & Andrews, G. (2007). Separating cognitive
capacity from knowledge: A new hypothesis. Trends Cogn Sci, 11,
236 –242. doi:10.1016/j.tics.2007.04.001
Haselgrove, M. (2010). Reasoning rats or associative animals? A common-
element analysis of the effects of additive and subadditive pretraining on
blocking. Journal of Experimental Psychology: Animal Behavior Pro-
cesses, 36, 296 –306. doi:10.1037/a0016603
Hashiya, K., & Kojima, S. (2001). Acquisition of auditory-visual inter-
modal matching to sample by a chimpanzee (pan troglodytes): compar-
ison with visual-visual intramodal matching. Animal Cognition, 4, 23–
239.
Jensen, A. R. (1989). The relationship between learning and intelligence.
Learning and Individual Differences, 1, 37– 62. doi:10.1016/1041-
6080(89)90009-5
Jensen, A. R. (1998). The g factor: The science of mental ability. West
Port, CT: Praeger.
Kaminski, J., Call, J., & Fischer, J. (2004). Word learning in a domestic
dog: Evidence for “fast mapping.” Science, 304, 1682–1683. doi:
10.1126/science.1097859
Kant, I. (1998). Critique of pure reason. New York, NY: Cambridge
University Press.
Kolata, S., Light, K., Grossman, H., Hale, G., & Matzel, L. D. (2007).
Selective attention is a primary determinant of the relationship between
working memory and general learning ability in outbred mice. Learning
and Memory, 14, 22–28. doi:10.1101/lm.408507
Kolata, S., Light, K., & Matzel, L. D. (2008). Domain-specific and
domain-general learning factors are expressed in genetically heteroge-
neous CD-1 mice. Intelligence, 36, 619 – 639. doi:10.1016/j.intell
.2007.12.001
Kolata, S., Light, K., Townsend, D. A., Hale, G., Grossman, H. C., &
Matzel, L. D. (2005). Variations in working memory capacity predict
individual differences in general learning abilities among genetically
diverse mice. Neurobiology of Learning and Memory, 84, 241–246.
doi:10.1016/j.nlm.2005.07.006
Kolata, S., Light, K., Wass, C. D., Colas-Zelin, D., Roy, D., & Matzel,
L. D. (2010). A dopaminergic gene cluster in the prefrontal cortex
predicts performance indicative of general intelligence in genetically
heterogeneous mice. PLoS. One, 5, e14036. doi:10.1371/journal
.pone.0014036
Lazareva, O. F., & Wasserman, E. A. (2006). Effect of stimulus orderabil-
ity and reinforcement history on transitive responding in pigeons. Be-
havioural Processes, 72, 161–172. doi:10.1016/j.beproc.2006.01.008
Light, K., Kolata, S., Denman-Brice, A., Zagalsky, R., & Matzel, L. D.
(2010). Working memory training promotes general cognitive abilities
in genetically heterogeneous mice. Current Biology, 20, 777–782. doi:
10.1016/j.cub.2010.02.034
Locurto, C., Fortin, E., & Sullivan, R. (2003). The structure of individual
differences in heterogeneous stock mice across problem types and mo-
tivational systems. Genes Brain Behav, 2, 40 –55. doi:10.1034/j.1601-
183X.2003.00006.x
MacWhinney, B. (1999). The emergence of language. Mahwah, NJ: Law-
rence Erlbaum.
Manktelow, K. (1999). Reasoning and thinking. East Essex, UK: Psychol-
ogy Press Ltd.
Matzel, L. D., Grossman, H., Light, K., Townsend, D. A., & Kolata, S.
(2008). Variations in age-related declines in general cognitive abilities of
Balb/C mice are associated with disparities in working memory span/
capacity and body weight. Learning and Memory, 15, 733–746. doi:
10.1101/lm.954808
Matzel, L. D., Han, Y. R., Grossman, H., Karnik, M. S., Patel, D., Scott,
N., et al. (2003). Individual differences in the expression of a “general”
learning ability in mice. Journal of Neuroscience, 23, 6423– 6433.
Matzel, L. D., Light, K. R., Wass, C., Colas-Zelin, D., Denman-Brice, A.,
Waddel, A. C., & Kolata, S. (2011). Longitudinal attentional engage-
ment rescues mice from age-related cognitive declines and cognitive
inflexibility. Learning and Memory, 18, 345–356. doi:10.1101/
lm.2034711
Matzel, L. D., & Kolata, S. (2010). Selective attention, working memory,
and animal intelligence. Neuroscience and Biobehavioral Reviews, 34,
23–30. doi:10.1016/j.neubiorev.2009.07.002
Matzel, L. D., Townsend, D. A., Grossman, H., Han, Y. R., Hale, G.,
Zappulla, M., et al. (2006). Exploration in outbred mice covaries with
15
REASONING AND GENERAL LEARNING
AQ: 11
AQ: 12
AQ: 13
AQ: 14
AQ: 15
AQ: 16
AQ: 17
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

general learning abilities irrespective of stress reactivity, emotionality
and physical attributes. Neurobiology of Learning and Memory, 82,
228 –240. doi:10.1016/j.nlm.2006.03.004
McGonigle, B. O., & Chalmers, M. (1977). Are monkeys logical? Nature,
246, 694 – 696. doi:10.1038/267694a0
Pilley, J. W., & Reid, A. K. (2011). Border collie comprehends object
names as verbal referents. Behavioural Processes, 86, 184 –195. doi:
10.1016/j.beproc.2010.11.007
Povinelli, D. J. (2000). Folk Physics for Apes. New York, NY: Oxford
University Press.
Rapp, P. R., Kansky, M. T., & Eichenbaum, H. (1996). Learning and
memory for hierarchical relationships in the monkey: Effects of aging.
Behavioral Neuroscience, 110, 887– 897. doi:10.1037/0735-
7044.110.5.887
Rashotte, M. E., O’Connel, J. M., & Djuric, V. J. (1985). Quantitative
Analysis of Behavior (Vol. 6): Foraging. Hillsdale, NJ: Erlbaum.
Roberts, W. A., & Phelps, M. T. (1994). Transitive inference in rats: A test
of the spatial coding hypothesis. Psychological Science, 5, 368 –374.
doi:10.1111/j.1467-9280.1994.tb00287.x
Sawaguchi, T., & P. S. Goldman-Rakic. (1991). D1 dopamine receptors in
prefrontal cortex: Involvement in working memory. Science, 251, 947–
950. doi:10.1126/science.1825731
Shettleworth, S. J. (2010). Clever animals and killjoy explanations in
comparative psychology. Trends Cogn Sci, 14, 477– 481. doi:10.1016/
j.tics.2010.07.002
Spearman, C. (1904). General intelligence, objectively determined and
measures. American Journal of Psychology, 15, 201–293. doi:10.2307/
1412107
Sub, H. M., Oberauer, K., Wittmann, W. W., Wilhelm, O., & Schulze, R.
(2002). Working memory capacity explains reasoning ability and a little
bit more. Intelligence, 30, 261–288. doi:10.1016/S0160-2896(01)
00100-3
Thurley, K., W. Senn, & H. R. Luscher. (2008). Dopamine increases the
gain of the input-output response of rat prefrontal pyramidal neurons.
Journal of Neurophysiology, 99, 2985–2997. doi:10.1152/jn.01098.2007
Unsworth, N., & Engle, R. W. (2000). On the division of short-term and
working memory: An examination of simple and complex span and their
relation to higher order abilities. Psychological Bulletin, 133, 1038 –
1066. doi:10.1037/0033-2909.133.6.1038
Unsworth, N., & Engle, R. W. (2007). On the division of short-term and
working memory: An examination of simple and complex span and their
relation to higher order abilities. Psychological Bulletin, 133, 1038 –
1066. doi:10.1037/0033-2909.133.6.1038
Williams, B. A., & Pearlberg, S. L. (2006). Learning of three-term con-
tingencies correlate with Raven scores, but not with measures of cog-
nitive processing. Intelligence, 34, 177–191. doi:10.1016/j.intell
.2005.03.007
Wynne, C. D. L. (1997). Pigeon transitive inference: Tests of simple
accounts of a complex performance. Behavioural Processes, 39, 95–112.
doi:10.1016/S0376-6357(96)00048-4
Received February 18, 2011
Revision received November 16, 2011
Accepted November 17, 2011 䡲
16 WASS ET AL.
AQ: 18
AQ: 19
tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
xppws S1 1/30/12 19:37 Art: 2011-0191

JOBNAME: AUTHOR QUERIES PAGE: 1 SESS: 1 OUTPUT: Mon Jan 30 19:37:57 2012
/tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
AQ1: Author: Please be sure to provide the name of the department(s) with which you and your
coauthors are affiliated at your respective institutes if you have not already done so. If you or
your coauthors are affiliated with an institute outside of the United States, please be sure to
provide the city, province (if applicable), and country in which the institute is based. If you
are affiliated with a governmental department, business, hospital, clinic, VA center, or other
nonuniversity-based institute, please provide the city and U.S. state (or the city, province, and
country) in which the institute is based.
AQ2: Author: Please limit to 5 or less keywords or phrases.
AQ3: Author: Editor feels this quote from Wall Street Journal should be cited.
AQ4: Author: Citation not listed in reference list. Please provide.
AQ5: Author: Citation not listed in reference list. Please check to see if all author names are list
and please provide.
AQ6: Author: Citation not listed in reference list. Please provide. Is 1853 a year or a page number?
AQ7: Author: Citation not listed in reference list. Please provide.
AQ8: Author: Should this be CS? Please check and give definition of US if needed.
AQ9: Author: Should this be CS? Please check and give definition of US if needed.
AQ10: Author: Should III be added to Lashley Maze?
AQ11: Author: Reference is not cited in text.
AQ12: Author: Please spell out journal name.
AQ13: Author: Reference is not cited in text.
AQ14: Author: Please spell out journal name.
AQ15: Author: Reference is not cited in text.
AQ16: Author: The APA does not allow the use of “et al.” in the reference list. For Author, Matzel,
L. D., Han, Y. R., Grossman, H., Karnik, M. S., Patel, D., Scott, N., et al. (2003), if author
names total 7, please provide all 7; if there are 8 or more, list the first 6 and add the last
author name after the ellipsis.
AUTHOR QUERIES
AUTHOR PLEASE ANSWER ALL QUERIES 1

JOBNAME: AUTHOR QUERIES PAGE: 2 SESS: 1 OUTPUT: Mon Jan 30 19:37:57 2012
/tapraid5/zfj-xan/zfj-xan/zfj00212/zfj2283d12z
AQ17: Author: The APA does not allow the use of “et al.” in the reference list. For Author, Matzel,
L. D., Townsend, D. A., Grossman, H., Han, Y. R., Hale, G., Zappulla, M., et al. (2006), if
author names total 7, please provide all 7; if there are 8 or more, list the first 6 and add the
last author name after the ellipsis.
AQ18: Author: Please spell out journal name.
AQ19: Author: Reference is not cited in text.
AUTHOR QUERIES
AUTHOR PLEASE ANSWER ALL QUERIES 2