Date post: | 21-Dec-2015 |
Category: |
Documents |
View: | 214 times |
Download: | 0 times |
The Cognitive Dog
Class 14: Continue on simple but reliable rules and responding to words...
Agenda
• Final paper
• Continuation on simple but reliable rules
• Word use
Where we have been...
• Effect of genes and development...
• social animal
• ability to imprint on other species
• later and much less intense fear period
Where we have been...
• The effect is an animal that has the potential to...
• selectively attend to members of species on which it has imprinted
• and level of fear is low enough (or missing) so that it doesn’t interfere with attention or learning.
Where have we been?
• Dogs as social animals do a ton of observational learning...
• Stimulus & local enhancement
• Observational conditioning
• Goal emulation (???)
• Simple forms of observational learning go a long way...
Where have we been?
• Dogs may use Simple but Reliable Enough rules to make sense of their world...
• We have seen lots of examples of behavior which is consistent with this hypothesis from invisible displacement to detours to observational learning...
Where have we been?
• We have looked at how dogs may use SBRE rules with respect to making sense of human behavior...
• Gestures as cues
• Attention as cues
• Words as cues
Where have we been?
• When dogs learn and use human cues to inform their behavioral choices, it is best to view it as a special case of...
• Interspecific Cue Learning (ICL), and
• Interspecific Cue Use (ICU)
• We use Interspecific Cue Learning (ICL) and Use (ICU) to emphasize...
• Interspecific: LGD that imprint on sheep, may selectively attend to sheep and not people and use sheep cues, and not human cues to guide their behavior...
• Cue learning and use: simple learning mechanism may suffice (not a special mechanism in of itself...)
So the big question...
• When dogs learn and use human cues to inform their behavioral choices is it,
• A special adaption in of itself?
• Or an emergent consequence of the interaction of
• imprinting & reduced fear?
• observational SBRE rule learning and use
Pongracz et al
Big idea...
• Dogs are conservative, they go with what worked in the past.
• what happens when what worked before doesn’t work any more?
• Observed behavior of human demonstrator can influence dog’s choice of behavior...
• what are they observing?
individual experience and social learning
Big idea...
• Previous success biases a dog’s choice of action even when an ‘obviously better’ choice of action is made available...
• dogs are conservative
• Dogs could use actions of a human demonstrator to guide their choice of action
Setup of experiment 1
Pongracz, P., A. Miklosi, et al. (2003). "Interaction between individual experience and social learning in dogs." Animal Behaviour 65(3): 595-603.
Door is never an option
Dogs first experience is
using door, which is subsequently
unavailable
Results from experiment 1
Pongracz, P., A. Miklosi, et al. (2003). "Interaction between individual experience and social learning in dogs." Animal Behaviour 65(3): 595.
Observing demonstrator
helps dog solve task
Previous experience with open
door seems to interfere
with choosing alternative
Results from experiment 1
Pongracz, P., A. Miklosi, et al. (2003). "Interaction between individual experience and social learning in dogs." Animal Behaviour 65(3): 595.
Dogs had a tough time
getting over the fact that the door was closed in the absence of a demonstrator
Experiment two...
• Dogs had varying previous experience with detouring
• 3 detour demonstration trial (dog gets to do detour)
• 1 detour demonstration (dog gets to do detour)
• Test trials with door open
• Trial 1 door opened while eyes covered
• Trials 2 and 3 dog sees food placed through open door
Results from experiment 1
Pongracz, P., A. Miklosi, et al. (2003). "Interaction between individual experience and social learning in dogs." Animal Behaviour 65(3): 595.
Chose to detour
Chose door
Dogs with more experience of
detouring persisted in detouring even when door was available
Results from experiment 2
Pongracz, P., A. Miklosi, et al. (2003). "Interaction between individual experience and social learning in dogs." Animal Behaviour 65(3): 595.
3 demonstrations
Dogs with a single demonstration were more focused on the door than those who
had three demonstrations
1 demonstration
Discussion
• Dogs tended to use first strategy that worked and would persist in using strategy even when a better one came along
• The more experience, the harder it was to change
• Socially acquired experiences given up slowly*
• Dogs did take advantage of watching a demonstrator with respect to taking a detour.
Discussion
• “Both of our experiments revealed the complex interaction between asocial (individual) and social learning that must be taken into account to understand how learning abilities in general contribute to increased fitness in animals. In experiment 1 social learning was advantageous in a situation where experience constrained the dogs’ behaviour. On the other hand in experiment 2, dogs facing a novel situation and exposed to socially provided information regarding access to the target were reluctant to change their behaviour, showing a preference for the more conservative (and socially learned) behaviour.”
Thoughts...
• Nice demonstration of observational learning and of conservative, to a fault, decision-making on the part of the dogs...
• I am less convinced that the way the learning occurred (social) meant that it persisted longer. The observation is conflated with the actual experience.
• They make a lot of this in the paper...
And yes...
• There may be an SBRE rule or two lurking here as well...
• go with what you know
• Is demonstration example of stimulus enhancement (end of fence) or local enhancement (places to wander)?
Follow-on study: Are dogs attending to paths or attending to corners (stimulus enhancement)???
Pongracz, P., A. Miklosi, et al. (2003). "Preference for Copying Unambiguous Demonstrations in Dogs (Canis familiaris)." Journal of Comparative Psychology 117(3): 337-343.
AmbiguousUnambiguous
Are dogs attending to paths or attending to corners (stimulus
enhancement)???
McKinley and Young
Masters of understatement
• “There has been one notable failure using this method, this involved a lar gibbon who attacked the model-rival during training sessions; this probably occurred due to the aggressive and territorial nature of this species”
• Lesson: know your animal, and don’t trust your advisor...
Big idea of model-rival...
• Animal observes trainer and a “model-rival”
• Trainer: “Can you see the SOCKS?” & hands toy to M-R
• Model-Rival: “Yes I can, thank-you for the SOCKS” & hands toy to trainer
• repeat...
• M/R is modeling desired behavior and rival for attention and/or possession of object.
Big idea of model-rival
• “fetch Mr. Squirrel” & receive a treat
• Q: in dog’s mind does the label “Mr. Squirrel” refer to the object or to the act of retrieving it and getting a cookie?
• A: who knows, but Pepperberg argues in traditional training, it is the latter, because the reward is extrinsic. This makes the label context-specific (i.e., tied to the specific action)
Big idea in model-rival
• In model-rival training, the animal’s reward for successfully performing the task is the object itself. The reward is intrinsic...
• Pepperberg argues that this makes it more likely that the label becomes associated with the object, and thus can be used in other contexts...
• “how many Mr. Squirrels are there?”
Big idea of model-rival...
• To keep dog’s interest, trainer & model-rival
• “speak in a highly animated way” [lots of motion??]
• “look at target object at all times”, but
• “voice direction and body postures directed toward dog”
• After 2 minutes, dog gets to try...
The test...
• After 2 minute training session...
• Trial
• Target object part of a group of 3 objects
• Dog told to “Go get the SOCKS”
• Trial ends when dog retrieves correct item, time noted.
• Note: order of retrieval doesn’t directly matter!!!
Results...
McKinley, S. and R. J. Young (2003). "The efficacy of the model-rival method when compared with operant conditioning for training domestic dogs to perform a retrieval-selection task." Applied Animal Behaviour Science 81(4): 357-365.
“We found no significant effects of training method on training times... or on trial times...”
Thoughts...
• Very cool example of using stimulus enhancement & maybe local enhancement to possibly help dog make association between label and object.
• interesting that dogs may be able to make this association
• Raises the whole question of intrinsic vs. extrinsic rewards
• Great to be exploring alternative approaches to training
Thoughts from Mr. Grinch...• We don’t know whether differences in training technique had any
impact on accuracy since this isn’t reported...
• why is article so silent on this point?
• What evidence is there for the following statement?
• “The m-r method resulted in the dog knowing what the object’s name was, whereas the operant method only resulted in the dogs knowing that retrieving the object results in a food reward”
Thoughts from Mr. SBRE
• Is there a Shallow But Reliable Enough Rule, or two, lurking here?
• “Stuff being handled by social partners must be good, and thus worth learning, remembering & finding...”
• “Pay attention to what social partners handle”
The slacker’s guide to the universe: simple but reliable rules
Words as cues...
Words as cues...
• The question isn’t whether pet dogs use words as cues to guide their behavior, but rather...
• distinguishing between use vs. understanding
• role of other cues in word use
• special ability to learn words(?)
Fellow: “the famous movie actor dog”
Fellow...
• 4 or 5 year-old GSD trained from early age to respond to words as cues...
• Owner claimed Fellow knew 400 or more words
• “specific objects, acts or places...”
Warden, C. J. and L. H. Warner (1928). "The sensory capacities and intelligence of dogs, with a report on the ability of the noted dog "Fellow" to respond to verbal stimuli." The Quarterly Review of Biology 3(1): 1-28.
Fellow...
• “He had quite successfully played, indeed, the usual roles alloted to his species in movie-melodrama–those of protecting the helpless and saving the drowning child...”
• “His claim to special attention lay in his accomplishment of responding to a large number of human words in some sense or other, and the problem was to determine in precisely what sense.”
The first test...
• Suite in the Pasadena Hotel in NYC...
• Mr. Herbert stationed in the bathroom and gave cues through a closed door...
• Warden, Warner and Fellow in the living room..
• Mr. Herbert varied the pitch, intensity and intonation of his voice when giving cues...
Fellow: the first test...
• “Although not perfect, the dog’s performance was on the whole quite satisfactory... it was evident that visual cues from Mr. Herbert were quite unnecessary to successful response in many cases at least, the essential stimuli being auditory in nature”
Fellow: the second test...
• Lab at Columbia in a more controlled situation (e.g., Mr. Herbert AND researchers hidden behind screens...)
• “repeat test to rule out possibility of visual cues”
• “secure data regarding total number of words to which the dog would thus respond”
• “test the ability to identify and retrieve a given object from among several upon command”
Fellow: the second test...
• A point worth noting:
• “The use of long test periods [one to two hours] was unfortunate inasmuch that Fellow often gave every indication of being weary of his task”
• “On all three days the command-performance test was given after Fellow had been working for an hour or more at the retrieving-objects test... and naturally he was not at his best on all occasions.”
Fellow: Type 1 and Type 2 Cues
• “It soon became apparent that certain commands could be carried out as perfectly under our test conditions as when the master was present but that others could not.”
• Two types of tasks...
• Type 1: performed correctly using acoustic cue only
• Type 2: requires both acoustic and visual cue to perform correctly
Type 1 cues
• Typically consist of “some movement ... in whatever place he happened to be at the moment”
• Does not require “very definite identification of object or place”
Warden, C. J. and L. H. Warner (1928). "The sensory capacities and intelligence of dogs, with a report on the ability of the noted dog "Fellow" to respond to verbal stimuli." The Quarterly Review of Biology 3(1): 1-28.
Type 2 cues
• “required the animal to identify and orient himself toward some object or place after the command had been given”
Warden, C. J. and L. H. Warner (1928). "The sensory capacities and intelligence of dogs, with a report on the ability of the noted dog "Fellow" to respond to verbal stimuli." The Quarterly Review of Biology 3(1): 1-28.
Tests of Type 2 Cues...
Mr. Herbert hidden behind screen
Mr. Herbert visible but giving
conflicting cues*
Warden, C. J. and L. H. Warner (1928). "The sensory capacities and intelligence of dogs, with a report on the ability of the noted dog "Fellow" to respond to verbal stimuli." The Quarterly Review of Biology 3(1): 1-28.
*Repeated with blindfold, and no difference in performance
W&W’s conclusions...
• “The fact that the dog did not fail at all on the more than 50 different commands of type 1... shows very clearly the ability of the dog to form associations between sounds, or verbal patterns and definite acts on his part”
• With respect to type 2 tasks: “it is only the object, or place in in connection with which the act is to be performed , and not the act itself that seemed to require visual cues”
What to take from this...
• Dogs can learn associations between words and actions
• non-referential actions (e.g., sit, lie-down, bark)
• referential actions (e.g., “get the ball”, “crate”...)
• they may rely to a greater or lesser extent on additional cues (e.g., visual) when available to help identify the referenced object or location
Scuppers & referential cues...
• I say “crate” facing the sink and Scuppers runs over to his crate under the stairs. Conclusion: Scuppers knows that “crate” refers to his crate.
• But, if I move the crate, show Scuppers where it is, and I say “crate”, Scuppers runs over to the location under the stairs where his crate used to be (even though he can see the crate in its new location.) Conclusion: “crate” refers to something, but just what is less clear...
Retrieving objects...
• Some referential actions seem easier than others for dogs to perform in the absence of visual cues...
• and this brings us to Rico, the over-achieving Border Collie who is especially adept at associating a novel word with the act of fetching a novel object...
Rico: the over-achieving border collie
Rico the wonder dog
• New Scientist (June 10, 2004).: “A word learning pet dog has given scientists clues that some animals may have the comprehension necessary for language, even though they cannot actually talk.”
What do you see?
1. Get the blue dinosaur (familiar)2. Get the red doll (familiar)3. Get the white bunny (novel)
Rico: the beginnings...
• ‘Introduced’ to fetching at 10 months.
• 3 objects in 3 locations
• Asked to retrieve an object, if correct, rewarded with food or play.
• Introduced to novel objects
• Presented to him and said name ‘2 or 3’ times
• Let him play with new item
• Integrated with other objects
Rico experiment
• Test 1: correctly retrieve known objects
• 20 sessions of 10 known objects each.
• In each session, Rico was asked to retrieve 2 randomly chosen objects from adjacent room.
• Reward: play or food if correct.
• 37 out of 40 correct
Rico Experiment
• Test 2: ability to fast map
• 10 sessions each introducing a novel object among 7 familiar objects
• 3 trials:
• Retrieve familiar object on trial 1, and maybe 2
• Asked to retrieve novel object using novel name on trial 2 or 3
• Correct on 7 out of 10 sessions
Rico Experiment
• Test 3: long term retention of novel label (4 weeks)
• 6 sessions each with 3 familiar, 3 totally novel, and a formerly novel object from test 2 (total of 9 objects)
• 1st trial asked to retrieve familiar object, on trial 2 or 3 asked to retrieve formerly novel object
• Correct on 3 out of 6 sessions (on par with 3 year olds.) Failure mode: retrieved one of the novel objects
Rico Experiment
• Test 4: Retention over 10 minutes
• 6 sessions each with a novel object.
• Ran test 2, wait 10 minutes, and then ran test 3.
• 4 out of 6 correct.
Kaminski et al’s claim
• “Rico’s performance can be decomposed into a set of simpler mechanisms...
• Acquisition of principle that objects have labels...
• A general learning mechanism, namely learning by exclusion...
• Ability to store that knowledge in memory”
Words as labels...
• To most of us the word “Sock” refers to category of object, independent of...
• color, material, size, smell, bruce’s old gym sock
• a sock placed on one’s ear is still a sock
• We can answer questions such as “how many socks”, “any socks”, “how many white socks” without fetching them
What do words mean to Rico?
• Bloom’s question
• Does “sock” refer to a category of object, or
• Does it refer to the act of fetching a particular object?
• Experiment doesn’t allow us to say either way
Bloom, P. (2004). Can a Dog Learn a Word? Science. 304: 1605-1606.
Do dogs use categories?
• The behavior of dogs seems to reflect an ability to recognize certain categories of things at least, e.g., dogs, cats...
• But we don’t know...
• what features dogs use to categorize those things that it can categorize (how does it recognize an individual cat as being a member of the cat category.)
• extent to which category can be separated from action
An experiment yet to be done...
Color or shape?
Train: “get the cone” Test: “get the cone”
To what features do they attend?
Train: “get the cone” Test: “get the cone”
An aside on bees...
• Bees learn to distinguish nectar sources based on
• time of day
• color of flower
• odor
• Do so in a predictable order too!!
Markman’s question...
• How does one distinguish “referential word use from mere association”?
• “My BC is smarter than your middle school student” may have hit too close to home :-)
• Her complaints...
• Effect of reinforcement for retrieving novel object
• No control for baseline novelty preference
Bruce’s response...
• Rico’s ability to “fast map” doesn’t appear to be in dispute. This ability is very cool. Some follow up questions though...
• Is it learned skill or gene X development
• Is Rico’s ability unique, unique to certain dogs, breeds?
• Does it reflect a general ability in dogs to learn acoustic patterns very quickly? If so, it is shared with other canids?
An aside on rats...
• Rats can learn to differentiate between spoken dutch and japanese sentences, but can’t do so if sentences are presented backward.
• Does this reflect an interplay between aspects of the acoustic structure of human language and auditory perception?
An aside on rats...
• “If languages have been shaped by, among other factors, constraints in the human auditory system, the ability to bootstrap linguistic regularities from low-level cues, such as rhythm, may represent the use... of an already-existing ability present in the mammalian auditory system”
Toro, J. M., J. B. Trobalon, et al. (2005). "Effects of Backward Speech and Speaker Variability in Language Discrimination by Rats." Journal of Experimental Psychology: Animal Behavior Processes 31(1): 95-100.
Bruce’s response...
• Since I am interested in what it says about dogs...
• That it works using mere association is a good thing, so reinforcement is not a problem.
• fetching itself may be reinforcing...
• bias to attend to novelty is cool: another SBRE rule that perhaps we should think about using more
Novelty preference
• Bias to attend to novel objects especially in familiar environment...
• Fetching as a self-rewarding behavior
• Rewarded for retrieving novel object, and fast maps cue to action of fetching that particular novel object.
• Aided and abetted by some bias toward mutual exclusion...
Being wicked smart about how smart to be...
Things to remember
Where we have been...
• Effect of genes and development...
• social animal
• ability to imprint on other species
• later and much less intense fear period
Where we have been...
• The effect is an animal that has the potential to...
• selectively attend to members of species on which it has imprinted
• and level of fear is low enough (or missing) so that it doesn’t interfere with attention or learning.
Where have we been?
• Dogs as social animals do a ton of observational learning...
• Stimulus & local enhancement
• Observational conditioning
• Goal emulation (???)
• Simple forms of observational learning go a long way...
Where have we been?
• Dogs may use Simple but Reliable Enough rules to make sense of their world...
• We have seen lots of examples of behavior which is consistent with this hypothesis from invisible displacement to detours to observational learning...
Where have we been?
• We have looked at how dogs may use SBRE rules with respect to making sense of human behavior...
• Gestures as cues
• Attention as cues
• Words as cues
Where have we been?
• When dogs learn and use human cues to inform their behavioral choices, it is best to view it as a special case of...
• Interspecific Cue Learning (ICL), and
• Interspecific Cue Use (ICU)
• We use Interspecific Cue Learning (ICL) and Use (ICU) to emphasize...
• Interspecific: LGD that imprint on sheep, may selectively attend to sheep and not people and use sheep cues, and not human cues to guide their behavior...
• Cue learning and use: simple learning mechanism may suffice (not a special mechanism in of itself...)
So the big question...
• When dogs learn and use human cues to inform their behavioral choices is it,
• A special adaption in of itself?
• Or an emergent consequence of the interaction of
• imprinting & reduced fear?
• observational SBRE rule learning and use
Dogs & Wolves...
• Dogs are descended from wolves
• Village dog hypothesis makes sense to me...
• humans as a new niche (15K BP)
• artificial selection may have occurred as well
• It is almost always a very bad idea to extrapolate from wolf behavior to understand dog behavior...
Just because it is there, doesn’t make
it adaptive
Evolved to allow other members of pack to locate dog when running through high prairie grass thus increasing chances dog wouldn’t get separated from pack and lose opportunity to reproduce
Nature always balances cost versus
benefit
He screws up and he goes hungry
He screws up and he dies
Epigenetic “evolution”
• Gottlieb...
• systematic change in developmental context can produce systematic change in behavioral and/or morphological trajectory, without changing “genes”
• what you see is the coaction of nature & nurture: each is “on tap” as opposed to “on top”
Nature & Nurture
• What one sees is a complex interplay of genes, development, developmental context, and learning...
• there is no such thing as “a dog”
• look to the parents for cues as to what a pup will be like
• pay attention to developmental periods...
• the ‘right’ socialization is key (no dog parks!!!)
The role of imprinting
• Dogs imprint early during their socialization period
• “Grow social connections in brain”
• Can imprint on a variety of species
• Leads to preferential attention to members of that species
• Exact mechanism is unknown...
Dogs & Perception
• Dogs perceive world very differently than us...
• contrast
• motion
• sound
• smell (but you knew that...)
• This all varies across breeds!!!!
Dogs see color differently...
Humans see: VIBGYOR Dogs see: VIBYYYR
Dana K. Vaughan, Ph.D., Dept. of Biology, University of Wisconsin, Oshkoshhttp://www.katiesbumpers.com/K9colorvision.htm
“The colors Green, Yellow, and Orange all look alike to dogs; but look different from Red and different from the various Blues and Purples. Dogs are very good at telling different shades of VIB apart. Finally, Blue-Green looks White to dogs.”
Simple but reliable rules...
• Your dog makes sense of its world via SBREs.
• Try to discover and pay attention to the SBREs your dog is learning and using.
• When in doubt assume your dog is using a SBRE
• I hope you have an appreciation for the power & elegance of using “simple rules & mechanisms” to make sense of the world...
Observational Learning...
• Stimulus-enhancement, local-enhancement and observational conditioning are all SBRE rules of how to learn from others.
• attend to objects that person/imprinted species is manipulating...
• may learn emotional reaction to other animals and objects from you (travels down the leash!)
Emotion: being ready to attend
• Later fear period in dogs and less intense secondary fear period may have made dogs “ready to learn”
• a small change with a huge effect
• When your dog doesn’t respond, it may because they are too stressed to respond.
• Threatening behavior often has its roots in fear
Dogs are emotional beings: the amygdala leads and the cortex follows...
Two paths to action: the low road (coarse features of situation) and the high road (complex analysis), and dogs are especially primed to react to the low road.
Find a dog with a good amygdala and soup of neurotransmitters
The power of classical conditioning
• Associate a good thing with a bad thing
• throw food on the ground, before, during, after.
• start out from a comfortable distance
• don’t ask anything of the dog
• if they react, move farther away
Emotion: self-rewarding behaviors...
• Broad view of “reward” and “reinforcement”
• Some motor patterns (typically appetitive motor patterns) may be self-motivating & self-rewarding...
• Terriers chase because “it feels good to chase”
• Typically skills that (a) may require practice to get right, and (b) even when done right, often fail...
• Just because there is no cheese, doesn’t mean there is no reinforcement signal!!!
Emotion: self-rewarding behavior
• It may be hard for your dog not to perform self-rewarding behaviors & even harder to break out of a self-rewarding behavior...
• The tricks...
• Be ahead of your dog
• Manage your dog’s environment (avoid rehearsals)
• Remember Premack: Use a self-rewarding behavior as the reward for a behavior that isn’t.
Emotion: Temperament
• Shy-Bold Continuum
• Playfulness
• Curiosity/fearlessness
• Predatory
• Social
• Threat/Reactivity
Emotion: Temperament
• Interplay of genes/development/developmental context/learning
• Temperament tests/breeder impressions are indicative, but not predictive (esp. if done before fear period)
• Mom & Dad are a good indicator
Learning & Training
• Your dog has to be emotionally ready to learn
• Training is about motivation, timing and consistency
• Positive reinforcement teaches dog what to do
• Punishment teaches dog what not to do
• potential for serious & unintended side-effects
Learning & Training
• Be a shaper not a lurer
• Let the dog find its way, but be there to guide it along the way.
• Be a splitter not a lumper
• Break down a behavior into its smallest bits and train those.
Learning & Training
• Your goal as a trainer should be to be the source of all things good and wonderful.
• Positive not permissive.
• Its ok to control access to resources
• Find reasons to reward rather than excuses to feed
• Find the thing your dog is nuts about and use it.
• Cookie cutters are for cookies, not dogs
Learning & Training
• Don’t read too much into the highs or the lows...
• Sometimes best to think of yourself as training someone else’s dog (takes the emotion out of it...)
• Remember what may seem easy to us, is a “triple lutz on national TV” to your little dog
• Be patient, calm, and confident and you will be rewarded
Learning & Training
• Is a journey to be enjoyed by both, rather than a means to a goal.
In the end...
• Dogs are special, but this does not imply special cognitive mechanisms and abilities, rather implies ‘special’ interplay of genes/development/learning that ‘we see but through a glass darkly’
A parting thought...
• “For the animal shall not be measured by man. In a world older and more complete than ours, they move finished and complete, ... living by voices we shall never hear. They are not brethren, they are not underlings: they are other nations, caught with ourselves in the net of life and time...” - Henry Beston
• Humility, awe and a passion to know more