Liz R. Kover
Professor Stanley Coren
LIF 515
December 15, 2014
DOGS
& COLOR CATEGORIZATION
Are
dogs able to recognize that two items belong in the same category based solely on
the sameness of their color? Comparing
and contrasting my experiment with Betty to John Pilley’s experiments with
Chaser, and others.
Chaser the Border collie, now famously
known as the smartest dog in the world, has achieved “academic” heights
previously presumed out of reach of the canine mind. Not only was John Pilley’s
groundbreaking research with Chaser a game-changer in the specific areas of
animal behavior he aimed to address, such as a dog’s ability to acquire
language skills and properly interpret syntax. But because language
comprehension is a definitive benchmark used to measure intelligence in human
beings, Chaser’s achievements are fueling an entire paradigm shift -- altering
not only what we know, but in fact the very nature of our inquiry into further
understanding. For ages, scientists have turned to non-human primates in
pursuit of insight into human intelligence. However, the dog -- once considered
a useless study subject in this realm –is now of ultimate interest as a species
whose convergent evolution with humans informs the development of both our
species’ cognitive make-up as socially cooperative (“pack”) animals. Thanks to Chaser, Rico, and other dogs
like them, the modern comparative framework puts a dog’s cognitive capacity
being on par with that of a human toddler. With this in mind, researchers are
simulating all kinds of cognitive psychology experiments that were originally conducted
with children, now using dogs as study subjects in place of two or three
year-old humans.
Many studies have established that
obvious (i.e. easy to observe and stable in time) cues such as color and shape
are used for categorization in infancy and early childhood (Nazzi & Gopnik,
2000). Therefore, using the resources I had at my disposal– namely a six
month-old black lab puppy named Betty, among a few other “supplies” – I set out
to watch a dog’s mind work with my own eyes. I designed an experiment that
combined elements from a few studies I found fascinating, including John Pilley’s,
to determine whether or not Betty could recognize color as a cue for
categorization. I “informally” hypothesized that given enough time and
practice, and a failsafe experimental model, it is likely that dogs could learn
to identify two objects as belonging to the same category based solely on the
sameness of their color. I reached my supposition based on the following
information revealed by previous studies:
(A) Contrary to popular belief, dogs
have and use color vision.
(B) Dogs are able to identify
individual objects by given names or labels.
(C) Dogs are capable of recognizing
objects whose names they have learned as independent from the commands with
which the objects are associated.
(D) Dogs are able to understand that
one word might apply to both an individual object, and a category within which
that object belongs.
(E) Dogs respond to non-verbal
communicative cues from humans, and infer their meaning.
(E-1) Dogs are able to translate two-dimensional
representations of objects as commands to retrieve the three-dimensional objects
they represent.
* *
*
My own study is a work in progress;
it will require more time and trials to complete, which means I do not have
conclusive data to report. Rather, I will compare my project to the studies on
which I based my hypothesis and experimental design; and will share the (inconclusive)
data I was able to glean from the work I did with Betty, which paints at least
part of an emerging picture, if not an entirely polished piece.
EXPERIMENTAL DESIGN & SET-UP
Using a clicker and training
treats, I rewarded Betty for indicating two different objects by the proper-noun
names I had given them. For simplicity’s sake, I named a yellow plastic puzzle
piece toy “Yellow”, and will refer to it herein as YO (yellow object); I named
a blue rubber bone “Blue”, and will refer to it herein as BO (blue object). I
first introduced (BO) by setting it on one of three equidistant “X”s I had drawn
on a large piece of white cardboard, and said to Betty, “Where’s BLUE”? When
Betty touched or nudged the object with her nose or paw, picked it up, or
obviously gestured in its direction by staring at it, I clicked and rewarded
the behavior. Then we did the same with the yellow object (YO). Next, we
repeated several trials during which both BO and YO were options. I made sure
to move the items from one X to another randomly between trials, so that Betty
was not simply conditioned to pick the object from one particular spot on the
piece of cardboard.
I had then planned to introduce two variables into the experiment: First, I would interchange the blue rubber bone with a square pad of Post-It notes that was virtually the same shade of blue, (herein I will refer to it as bp for “blue paper”), and interchange the yellow plastic puzzle piece toy with a pad of Post-It notes that was virtually the same shade of yellow (herein referred to as yp for “yellow paper”). I would repeatedly present her with two items at once in one of three randomly-ordered contextual set-ups: Either (a) both original colored objects, or BO & YO, (b) both colored squares of paper, or bp & yp, or (c) one object and its oppositely-colored square of paper, or BO & yp/ bp & YO. *Had Betty chosen a colored pad of paper when presented with one as an option, she would’ve been choosing it by virtue of its color alone, as the colored pads of paper shared no other similarities with the original objects besides their color. They were different in texture, size, shape, and weight; and Betty wouldn’t have seen them before they were brought into the experimental context. [*We did a brief few trials using this set-up, which you’ll see at the very end of the video on YouTube. But we didn’t complete nearly enough trials to obtain significant results one way or another.]
Next, I would have the original
blue and yellow objects set up on the X’s, as before. But this time I would
hold up either the blue Post-Its or the yellow ones, and see if Betty would
infer that by showing her the paper, I wanted her to select the object that was
“the same” as the paper. I will go further into the implications of what this
test would reveal shortly.
DOGS HAVE AND USE COLOR VISION
Dogs’ eyesight directly affects the
efficiency with which they navigate the world around them. Consider the keen
visual acuity it takes for a dog to hunt, or in fact become a vehicle of sight
for a person who’s lacking it. While dogs’ visual prowess primarily lies in
capturing fast-moving objects and/or those skittering across the landscape in low-light,
studies have clearly documented that dogs possess and use color vision as well.
Human eyes contain three types of color-receptive cones; therefore we are able
to perceive a richer, more saturated, and wider color palette than dogs, whose
eyes are equipped with only two classes of cone photopigment. In other words,
humans have trichromatic vision, while dogs’ is dichromatic. The visible
spectrum in dogs is divided into two hues: one in the violet and blue-violet
range, which is presumably seen as blue by dogs, and one in the
greenish-yellow, yellow and red range, which is probably seen by dogs as yellow
(Miller and Murphy, 1995).
When designing my experiment, I had
to be certain that Betty would not only be able to see the individual colored
objects I was using, but also discriminate the two colors from one another. In
other words, had I chosen objects that were green and yellow, there likely
would not have been enough contrast for Betty to perceive a difference. Or had
I chosen a red object and a green one, Betty would likely have only seen two
shades of gray with little variation in between. I chose the colors I did
because wavelengths at the two ends of the visible spectrum in dogs’ visible
field – blue at one end and yellow at the other end – likely provide the most
saturated and contrasted colors (Miller and Murphy, 1995).
In addition to merely having the
anatomical capacity for color vision, studies show that dogs make use of the
potential therein implied. Recent research out of the Royal Society has shown
that dogs use chromaticity over brightness to glean information from colors in
their environment. Scientists tested dogs’ preference when choosing test
stimuli that differed in both hue and brightness. Incidentally, these
researchers used yellow and blue for their experiment like I did, but they used
two highly contrasted yellows, one bright and one dark, and did the same with
dark and light blues. From their results, the team concluded that under natural
photopic lighting conditions, color information may be predominant even for
animals that possess only two spectral types of cone photoreceptors (Kasparson,
et al. 2013).
WE KNOW THAT DOGS ARE ABLE TO LEARN OBJECTS BY NAME
Chaser the brilliant Border Collie ultimately
learned to identify 1,038 objects by name over a three-year period. Assuming she
and her person, John Pilley, trained 350 out of 365 days per year, approximately
4.5 hours per day, then one could assume they dedicated an average of 9.5 hours
of training time to each object. Betty and I were able to dedicate approximately
1/5th of the necessary amount of time and practice it would take for
us to achieve comparable results. After 100 trials of having Betty choose YO or
BO, she was barely choosing correctly above chance, and seemed to be guessing
for the most part about which object was which. When both objects were present,
Betty chose correctly 56 times, and incorrectly 44 times. Had we had more time,
and had she learned to identify each object in reliable “Chaser fashion”, our
experiment would’ve moved beyond the theoretical and hypothetical stages. But
alas, it did not. Betty’s age, breed, and level of previous generalized
training must also be considered. Betty is a six month-old Labrador retriever
in training to become a service dog for a child with autism. She is an
intelligent dog, by my subjective account, but I wonder how the experiment
might’ve gone differently if she were a Border Collie or Australian Shepherd,
for example. I understand this is a variable in all canine cognition
experiments that involve participants of only one breed, and especially for
those involving only one participant. Had Betty had more time to practice
identifying objects by name, I am sure we could’ve gotten to the “meat” of the
color categorization hypothesis more quickly. As it stands, we have no
conclusive results to report just yet, though we plan to continue working on it.
DOGS RECOGNIZE OBJECTS AS INDEPENDENT FROM THE COMMANDS WITH WHICH THEY
ARE ASSOCIATED
Critics of the Rico
language-learning study questioned whether or not Rico understood that the
phrase “fetch sock” represented two independent morphemes – that objects are
independent in meaning from the activity requested involving that object (Pilley
and Reid, 2011). Pilley countered this criticism by conducting “Experiment 2”
with Chaser, to determine whether Chaser treats the name of an object
independent in meaning from the command given in reference to it (Pilley and
Reid, 2011). Chaser was asked to act upon each of three named objects, “Lips”,
“ABC” and “Lamb”, whose proper-noun names she had previously learned. While
Chaser’s action of “taking”, “pawing”, or “nosing” an object when commanded to
do so determined that she did, indeed, understand the action and object as
independent from one another, what I did in my experiment with Betty indicated
something altogether different, but still important and interesting.
I started out asking Betty to “get”
(pick up in her mouth), “go get” (go a distance and pick up in her mouth), or
“touch” (with nose or paw) either the blue object or the yellow object during
the “learning the proper-noun name phase” of our experiment. I also used
commands such as “drop blue”, or “give yellow”, while we were just “playing
around” with the toys, so as to identify the objects by their given names as much
as possible, thereby reinforcing the information in Betty’s mind. During this
phase, I was impressed at Betty’s ability to discriminate between commands that
required subtly different physical actions on her part – such as “touch” and
“get” - especially when no context change occurred in between trials. That
being said, after a while, for the sake of keeping it simple, I omitted the use
of different action commands, and replaced them with just one word (or its
conjunctive phrase): “WHERE’S or WHERE IS”. So I would say “Where’s blue?” or
“Where’s yellow”. Interestingly, Betty would do one of five things when
commanded only with “Where’s” or “Where is”. She would either (a) touch the
item with her paw; (b) pick up the item with her mouth; (c) nudge the item with
her nose; (d) nod her head quickly in the direction of the item; or (e) look
obviously in the direction of the item, staring at it for a couple of seconds. To
my mind, this implied that Betty was making an inference as to what I wanted
from her. I had not previously used the command “Where is” in any other
context, so she figured out on her own that all five of the aforementioned
actions accomplished what I “implied” I wanted from her. She figured out that I
wanted her to indicate, in one way or another, which object I was asking her
about. I found this fascinating, especially in concert with all the other
research findings that support social and cooperative learning in both species
as related to our convergent evolutionary paths. This was yet another example
of how dogs have evolved to read subtle human cues in novel situations, in
which there has been no prior training or experience.
DOGS ARE ABLE TO UNDERSTAND THAT ONE WORD MIGHT APPLY TO BOTH AN
INDIVIDUAL OBJECT AND A CATEGORY TO WHICH THAT OBJECT BELONGS
Another elemental criticism of the
Rico study that John Pilley aimed to counter in his study with Chaser had to do
with distinguishing between proper-noun names given to objects, and common
nouns that represented categories. For instance, if Pilley had named a stuffed
round object of Chaser’s “Ball” (let’s call the character “Ball Jones”), that
would be distinctly different than him picking up a ball, and calling it “a
ball”. In other words, Ball (Jones) with a capital “B” refers to an individual,
named object, whereas “ball” with a small “b” refers to an item that represents
a category of objects that fit the physical and functional description of what
a ball is: it is spherical, it rolls, it bounces, etc. Pilley notes that
categorization in animals has been widely studied in animal cognition in the
last few decades, and a few studies have in fact demonstrated that dogs can form
categories (Pilley and Reid, 2011).
Chaser learned the categories “ball”,
“frisbee”, and “toy”. She could determine the sameness of balls and other balls,
frisbees and other frisbees by their shapes; while she seemed to determine that
toys – categorically – were items that served a particular function; that is, toys
were things Chaser was allowed to play with. She even distinguished her toys
from other objects that were similar in the house, but which were non-toys. For
example, one of Chaser’s toys was a sock (let’s call him Sock McGillicutty,
Sock for short), but then there were other socks in the house that were not for
her to play with, and which she left alone, never confusing them with her toy
sock.
In my experiment with Betty, I
aimed to see if she would be able to categorize objects when the only likeness
they shared was their color. In my experiment, I used the proper-noun names,
Yellow with capital “Y” for the plastic puzzle piece and Blue with a capital
“B” for the rubber bone. (For continuity’s sake, we could call these two
characters “Yellow Johnson” and “Blue McGee”). In this case the categorical
symbols were not common nouns, but rather adjectives. In other words, the
yellow and blue post-it note pads represented the color categories (yellow with
a small “y” and blue with a small “b”). Say I laid out the blue bone, “Blue”,
and the yellow Post-It notepad, and asked Betty, “Where’s yellow?” In this case
I would not be asking for “Yellow Johnson”, the proper-noun-named individual.
Rather, I would be asking her to show me which item before her belonged in the
category of “yellow things”.
Were Betty and I to have made it further
into our experiment, we would have practiced generalizing and discriminating,
as Pilley did with Chaser. I would’ve laid out several yellow objects, and
several objects that were not yellow
(and had no yellow in them), and asked Betty to bring me the yellow objects.
Or, I would ask her to bring me a (pink object, for example), leaving one pink
object in a group of yellow objects, to see if she could determine which one
was pink based solely on the fact that it was not yellow.
DOGS RESPOND TO NON-VERBAL COMMUNICATIVE CUES FROM HUMANS, AND INFER
THEIR MEANING
Dogs have been shown to infer from
two-dimensional images that the experimenter wants them to choose an object
that the two-dimensional image depicts. For this reason, I imagine Betty
would/will be able to infer that when I hold up the yellow or blue Post-It
notepads, I am asking her to choose the colored object whose color corresponds
with whichever notepad I am holding up.
In 2009, the Dog cognition “dream
team” at Cambridge University and Max-Planck Institute for Evolutionary
Anthropology published a paper called Domestic
Dogs Comprehend Human Communication With Iconic Signs. In it, the authors
examined dogs’ ability to infer the communicative intentions of humans when
given an abstract, non-verbal cue to fetch a referent object. In their paper,
they point out that infants are able to make appropriate inferences and
behavioral responses to adults’ communicative intentions (Kaminski et al.,
2009). Evolutionarily speaking, this adaptation allows infants to benefit from
adults’ guidance prior to the development of language comprehension. By the
time a baby has reached two to three years old, he or she is able to infer from
being shown a replica or photo of an object, that the adult showing the baby
the icon wants for him or her to produce the real thing. Due to the emerging acceptance
of a dog’s cognitive capacity being on par with that of a human toddler, it
made sense for the dogs to have excelled at this particular test. That said,
the notion that dogs are capable of handling the dual representation of an
object that is both a replica of another object, and an object in its own
right, is rather mind-boggling.
In this study, researchers
investigated the skills of domestic dogs – some language-trained and some not –
to make appropriate inferences and behavioral responses to human communicative
intentions as expressed in their use of iconic signs, specifically physical
replicas and photographs (Kaminski et al., 2009). Again, test subjects chosen
for this study were Border Collies (one of them being Rico), who had
overwhelming success with fetching objects when they were shown – in a
“communicative context” either photographs of said objects, or smaller,
physical replicas of them. The dogs even succeeded at choosing objects
represented by the photographs, over identical copies of the photographs, when both were options. This further
supports the truism that dogs infer meaning from what humans communicate to
them, even when there is a level or more of abstraction involved.
This concept played a role in my
experiment with Betty, however it translated differently than in the Kaminski
study. When I interchanged the
colored Post-It note pads with the plastic and rubber toys, and asked Betty,
”Where’s (Y)ellow/(y)ellow” (or (B)lue/(b)lue), I was asking her to interpret
and understand the pad from three different perspectives: The (yellow) pad was an
object in its own right (though unnamed and unfamiliar to her); it was a
representation of the named item, “Yellow”; and by virtue of its color, it
represented the category into which both it and its referent object (YO)
belonged .
Had we reached the point in our
experiment where I held up the Post-It pad silently suggesting that Betty show
me its match, (and were Betty to have succeeded at this task), it would show
what Kaminski’s study showed: that dogs’ skills in using human forms of
communication are flexible as they generalize immediately to a new perceptual
modality (vision vs. audition) (Kaminski, et al., 2009).
DOGS ARE ABLE TO CATEGORIZE ITEMS WITH OR WITHOUT HUMAN CUES
Friederike Range and his team at the
University of Vienna, reasoned that “although domestic dogs are on the brink to
become one of the model animals in animal psychology, their categorization
abilities are unknown”, and that “this is probably largely due to the absence
of an adequate method for testing dogs’ ability to discriminate between large
sets of pictures in the absence of human cueing” (Range, et al. 2007). So,
while Kaminski looked at dogs’ ability to draw conclusions about the task at
hand based on non-verbal communication between the dogs and researchers, Range
wanted to zero in on dogs’ ability to complete categorization tasks without
outside help.
As of 2007, when this study was
carried out, it was only one of two known categorization studies using dogs as
subjects, the other being an acoustic stimuli categorization test to see
whether or not dogs could discriminate dog from non-dog sounds. As was true for
my experiment with Betty, the authors of this study had to fist take into
account evidence for dogs’ ability to see in color. They determined – as did I
– that although dogs have reduced color perception, they expected no severe
physiological limitations of the dog’s ability to classify color (photographs),
provided the category-specific aspects were not restricted to shades of red or
very tiny fragments of the pictures (Range et al., 2007). Originally, I was using a third object
in my experiment: along with the blue bone and yellow puzzle piece, I
introduced a hot pink collar to Betty as well. I reasoned that, even if she saw
the pink collar as a shade of gray, it would still be different enough from
either the yellow or blue objects that she would put it in a category all its
own. I later removed it though, because we weren’t progressing very quickly
with two items, so I thought I would make it easier by eliminating a third.
The question most relevant to the
Range study and my own, was if the dogs would be able to distinguish between
category-relevant and category-irrelevant features (Range, et al., 2007). When I asked of Betty that she identify
the yellow Post-It note pad as “(y)ellow”, after having been taught a plastic
object with which the notepad shared one feature, its color, was called “(Y)ellow”,
I was looking to see if she would practice a category-specific strategy. A category-specific
strategy would require the subject to extract and combine the features common
to most (or maybe even all) instances of a class and then react in the same way
to all stimuli possessing those features (Cook et al., 1991). In other words,
even though Betty had learned the plastic object by its name, “Yellow”, she
would’ve also chosen the yellow notepad when I asked her to show me “yellow”.
All of the dogs in Range’s study
eventually mastered the task of classifying photographs according to the presence
or absence of dogs in them, and could transfer this practice to novel and even
contradictory stimuli. Betty and I will need more time and practice before we
can report on her ability to classify yellow or blue Post-Its as belonging to
the same category as congruently-colored toy objects.
WE ARE THE SAME AND WE ARE NOT THE SAME
While achieving significant results
showing that dogs’ cognitive capabilities are very similar to humans is exciting,
results that disprove these types of
hypotheses are just as valid and notable. While it is important to
recognize similarities in human and canine minds, it is just as important to
appreciate dogs and human beings, respectively, as unique and different beasts
altogether. With results pending on my own experiment, the reality could go
either way. Dogs may have and use color vision, and they may utilize it in ways
similar to human beings when categorizing information in their environment. But
then again, they may not.
A recent study on shape bias in dogs,
for example, questioned whether or not dogs (in this case a five year-old
Border collie named Gable) use an object’s shape to generalize it to objects in
the same category. Using an experimental paradigm originally established to
examine shape bias in children, researchers showed that when (Gable was)
briefly familiarized with word-object mappings the dog did not generalize
object names to object shape but he did to object size. Another experiment
showed that when familiarized with a word-object mapping for a longer period of
time the dog tended to generalize the word to objects with the same texture.
These results show that the dog tested did not display human-like word
comprehension, but word generalization and word reference development of a
qualitatively different nature compared to humans (van der Zee et al., 2012).
Again, results that illuminate the differences between the human mind and that
of Caninekind are equally valuable to
an overall understanding of who both we – and they – are.
* *
*
Father of evolution Charles Darwin
said, “The difference in mind between man and the higher animals, great as it
is, certainly is one of degree and not of kind”. The nonhuman primate mind has
long been considered the closest “in kind” to humans’. But today, dogs are
ushering in a brand new era of cognitive exploration, leading us to look at
animal behavior as related to social learning, cooperation, and communication
between and among species. During the domestication process, dogs developed
innate sensibilities to human emotional states, communicative cues, and comprehension
of verbal language. The results of Chaser’s and other dogs’ studies are as much
an existential statement on the meaning of intelligence – and the existential
quandary of being - as they are a
declaration of these particular animals’ beyond-impressive ability to learn,
remember, and categorize objects. Rather than asking merely “how intelligent
are dogs?”, Chaser, and other canine pioneers have us asking, How might various manifestations of canine
and human intelligence have developed in the scope of our convergent evolution
as interspecies partners? And, putting to good use what we’re learning from
dogs in research settings we might ask, How
can we use our findings to better communicate with the dogs in our lives whom
we love so much, and seek so desperately to understand?
ADDITIONAL INFORMATION
You
can watch video of my experiment with Betty here: http://youtu.be/hIV8Y2L_G_k?list=UUeZtG7Bv7VQhtERlQ_tPs7g
References
Pilley, J.W., Reid, A.K., (2011) Border collie comprehends object names as
verbal referents. Behavioral Processes. doi: 10.1016/j.beproc.2010.11.007
Miller, P.E. and Murphy, C.J., Vision in Dogs. (1995) Journal of the
American Veterinary Medical Association. Vol 207, No. 12, pp. 1623-1634.
Kasparson, A.A., Badridze, J. and
Maximov, V.V., (2013) Colour cues proved
to be more informative for dogs than brightness. Proceedings of the Royal
Society B 280: 20131356. http://dx.doi.org/10.1098/rspb.2013.1356
Kaminski, J., Tempelmann, S., Call,
J. and Tomasello, M. (2009) Domestic dogs
comprehend human communication with iconic signs. Developmental Science. Vol
12, No 6, pp. 831-837. DOI: 10.1111/j. 1467-7687.2009.00815.x
Range, F., Aust, U., Steurer, M.
and Huber, L. (2007) Visual
categorization of natural stimuli by domestic dogs. Animal Cognition. DOI:
10.1007/s10071-007-0123-2
Van der Zee, E., Zulch, H. and
Mills, D. (2012) Word Generalization by a
Dog (Canis familiaris): Is Shape Important? PLoS ONE 7(11): e49382. DOI: 10.1371/journal.pone.oo49382
Nazzi, T. and Gopnik, A. (2001) Linguistic and cognitive abilities in
infancy: when does language become a tool for categorization? Cognition. Vol
80. B11-B20