Greg Detre
February 16, 2003
After
showing that dogs can out-perform chimps at making use of human social cues,
such as gaze direction, pointing, touching the object, or other novel or
arbitrary cues (9/11 dogs vs 2/11 chimps using social information to find
hidden food in their experiment), Hare et al. set out to find out which of the following
hypotheses best explain domestic dogs� skill in using human social cues:
canid generalization hypothesis: predicts that many canids
(especially wolves) should perform at least as well as dogs on social tasks, as
been found previously with non-social tasks
human exposure hypothesis: variation in individual dogs�
experience with humans will be associated with variation in the task
performance, and that young dogs should therefore have relatively poor skills
domestication hypothesis: selection pressure on dogs during
the process of domestication for specific skills of social cognition and
communication with humans
Their
approach is to compare:
a) adult human-raised wolves and dogs
on both a social and a non-social task;
b) puppies of varies ages and amounts
of exposure to humans in an object choice task.
In a) the
dogs consistently out-performed the wolves using combinations of gazing,
pointing and touching cues, although they did not perform above chance in the
control condition. This cannot simply be attributed to domestic dogs� higher
intelligence or aptitude for all human-guided tasks, since dogs and wolves
performed equally well on a non-social human-guided memory task.
With regard
to b), there was no difference in performance in the same social task between
two sets of puppies differing in exposure to humans during rearing (raised by
humans from 8 weeks vs living in a kennel with only a few minutes of human
exposure per day), nor was any effect of age detected using three age groups
(9-12 weeks, 13-16 weeks, 17-24 weeks).
I would be
interested to see a follow-up experiment designed to test whether the
difference between domestic dogs and wolves is a result mainly of an evolved
ability to map human movements onto a canine body plan. After all, we might expect
I felt that
Hare et al. provided powerful evidence that domestic dogs have recently evolved
to take advantage of human social cues in some ways. However, given that the
experimenters make the reasonable assumption that wolves would use social cues
from conspecifics and quarry in the wild, I would be interested to see a
follow-up experiment with wolves designed to test whether they are better at
using conspecific social cues than domestic dogs. After all, if the only thing
that dogs evolved was an ability to map human movements onto their own motor
plan, that alone would give them a tremendous advantage in this sort of
experiment.
In fact,
it�s difficult to imagine what other features domestic dogs could have evolved
within a relatively short time (the last 100,00 years or so) to make them so
much more successful at the task than chimps. Their ability to make use of
novel or arbitrary cues is especially interesting in this regard. This is the
sort of wide-ranging ability that presumably underpins a full-blown theory of
mind.
Slabbert
& Rasa�s aim was to determine how well working dog pups can acquire a
maternal learned behaviour pattern during early ontogeny through observation
and retain this information in later life. They divided 95 German Shepherd pups
into four groups, according to whether or not their mother was trained to
retrieve packets of hidden narcotics, and whether they spent 6 weeks (as is
normal in dog schools, so that the pups can bond with their human trainers) or
12 weeks (as is more natural for dogs) living with their mothers.
Untrained mother |
Trained mother |
||
6 weeks |
12 weeks |
6 weeks |
12 weeks |
Group I |
Group II |
Group III |
Group IV |
Their finding that simply �seeing its mother perform the learned
behaviour and be praised for it a total of 14 times with a 6 week period in
early ontogeny� was sufficient to make the success rate of suitable pups in
Group IV 85% as opposed to 19% in all the other groups is astonishing.
The other
three groups controlled for simply being in contact with the mother for longer
(Group II) and for inherited aptitude (Group III), which are the two main
factors that one might consider relevant.
Slabbert
& Rasa seem to have comprehensively demonstrated the positive effects on
pups� learning of being reared in contact with their mother and watching them
being rewarded. However, I would have liked to have seen them:
a) consider what in particular about
the experience facilitates it;
b) speculate about how to further
improve upon the technique used in the experiment;
c) speculate about how this rapid
learning might work.
With regard
to a) and b), it would be illuminating to run further experiments,
investigating:
1) the effect of demonstrations by the
mother as opposed to some other non-maternal trained dog, or even a human
trainer, to see whether the puppies pay particular attention to their mothers
2) whether you could facilitate the
puppies� learning even further by rewarding the mother or the pup on completion
of the task, or by actively involving the pup in the task � presumably dog
schools have experimented with trying to teach puppies at this early a stage in
their lives
3) whether the puppies were learning
simply by having their attention focused on the smell of the narcotics
(associated with the reward given to their mother), whether they were learning
something of the nature of the task (finding hidden parcels and bringing them
back to the trainer), or whether they were learning something about the type of
places where trainers hide the parcels
4) whether the pups would learn even if
the mother wasn�t rewarded for completion of the task
Pongracz et al. used a V-shaped chain-link detour fence with a favourite toy or food on the opposite side from the starting point. This problem is suitable because it is of an appropriate level of difficulty, has two solutions (either left or right), and can be topographically reversed (inside vs outside of the V-shape) to provide a new problem of similar type. In contrast to the Slabbert & Rasa paper, the emphasis here is on dogs� ability to learn a new task from human demonstrators.
Interestingly, the dogs did not find the Inward/Outward detour problems to be symmetrically difficult. Dogs in the Inward detour group did not show significant improvement in latency (the time taken between being released from the leash and taking the object in their mouth), whereas the Outward detour problem was mastered much more easily.
In Experiment 2, Pongracz et al. compared Experimenter Demonstration, Owner Demonstration and No Demonstration groups to see whether observing a human making a detour might enhance learning in na� dogs. While dogs did not show significant improvement without demonstration over three trials, a single human demonstration shorteened latency significantly. The fact that this effect worked with both owner and experimenter demonstrations indicates a general predisposition to copy human behaviour. Interestingly though, the dogs did not copy the demonstrator�s exact route, but clung instead to the direction of their own first successful trial.