Friday, December 21, 2012

Wittgenstein, Sheep Thieves, and the Dog Who Knew 1,000 Words

Ludwig Wittgenstein begins the Philosophical Investigations by quoting St. Augustine’s description of how he had learned to speak, where adults would point out objects and give them names, so that Augustine “grasped that the thing was signified by the sound they uttered.” This is, Wittgenstein says, “the idea of a language more primitive than ours.” He explains how this language would work:

“Let us imagine a language for which the description given by Augustine is right: the language is meant to serve for communication between builder A and an assistant B.  A is building with building stones: there are blocks, pillars, slabs and beams.  B has to pass him the stones and to do so in the order in which A needs them.  For this purpose they make use of a language consisting of the words ‘block’, ‘pillar’, ‘slab’, ‘beam’.  A calls them out; B brings the stone which he has learnt to bring at such-and-such a call. — Conceive of this as a complete primitive language.”

This primitive language is fairly similar to how we train dogs and then expect them to behave. (As Kingsbury Parker pointed out to me, Wittgenstein's description is closer to the way a surgeon works with nurses in the operating room than how masons actually work.)  Dogs fetch something for us after we have identified what we want fetched. It must also be noted that dogs communicate with us by commands, which are not verbal but may be vocal.  If my dog needs to relieve herself, she paws a bell we keep on our front door.  If she wants to take a walk, she brings her leash.  Some dogs will bark near a door to make make such requests.   

Eric Krieger, in reviewing a draft of this blog, pointed out to me that an initial command might also be the trigger a series of actions from both sides.  He describes a game he plays with Kenai, his Mackenzie River Husky:

“In the evenings about an hour before Kenai sacks out for the night we like to have a play time. She has a few toys that she's played with since puppyhood. A stuffed monkey and a couple of nylon-bone keys (chew toys). When a say, "get your monkey, or get the key" she runs around the house, finds the requested object, runs to the bedroom door and takes about a seven foot leap onto the bed, where we play tug.”

Of course, times, places and circumstances may also vary how either the handler or the dog will act and react.

The Sheep Thieves

How much complexity can be introduced into this shared language given the dog’s mental capacity?  Consider the following passage from Edward C. Ash (1927, p. 271):

Sheep Dog (Hamilton Smith, 1840)
“[Sydenham Edwards] tells a story of a sheep-stealer visiting a flock with the presumed intention to purchase.  His dog went with him, and on this inspection he secretly gave the dog the hint as to which of the sheep he wanted, to the number, we read, of ten or twelve, in a flock of some hundreds.  That night, from a distance of as many as twelve miles, he would send back the dog, who picked out the individual sheep, separated them from the flock, and brought them to his master.” 

Additional detail might be necessary to say whether this could really have happened, which Ash doubts.  How did the thief “hint” at the sheep he wanted?  Did he point? Touch the sheep he wanted to steal?  Were they the largest sheep in the group, or the smallest?  Ewes of a certain size?  The black-faced sheep?  The sheep on one hill as opposed to another?  Leaving aside the legendary flavor of the story, recent research may say something about the limits of a dog’s ability to identify specific objects of a master’s interest and follow a command with respect to those objects.  Research is also saying more and more about how dogs learn such skills.  

How Many Words Can a Dog Learn?

In 2004, three scientists at the Max-Planck Institute for Evolutionary Anthropology in Leipzig, Juliane Kaminski, Josep Call, and Julia Fischer (Kaminski et al.), published a paper on Rico, a Border collie, whose owners had taught the names of over 200 items, mostly children’s toys and balls that he could fetch on command. 

Two researchers at the Konrad Lorenz Institute in Austria, Ulrike G. Griebel and D. Kimbrough Oller (Griebel and Oller, 2012) tested a Yorkshire terrier named Bailey that had been taught the names of about 120 toys by her owner, Noreen Yuki, and would fetch the correct one if commanded to do so.  Bailey’s toys often had names of two or more words:  Frosty the Snowman, Red Rose, Football, Nemo, Ladybug, Green Bone, Rocco Raccoon, Ozzy Ostrich, Beatrice Bat, Heidi Hippo, Iggi Iguana, Long Legged Leopard, Victor Vulture, Suzie Sunshine, Louie Lobster, Cat in the Hat, Mushy Mushroom, Twinkle Twinkle Little Star, and Rudolph the Rednosed Reindeer.  Griebel and Oller also found that Bailey did well in fetching objects regardless of whether the command was given by her master or someone else, and also did well if someone with a German accent gave a command. 

The record, at least of dogs described in the scientific literature, appears to be held by Chaser, a female Border collie that, over a period of three years, was taught the names of 1022 objects, over 800 cloth animals, 116 balls, 26 frisbees, and about 100 plastic items.  Chaser lived in the home of the researchers, and worked with them four to five hours a day on learning the names of the objects.  The training stopped, according to John W. Pilley and Alliston K. Reid (2011), not because Chaser had reached some cognitive limit but because they could no longer invest four to five hours a day training her. 

Commands Relative to Objects

Wittgenstein, elaborating on his example of the two builders, noted that in his simple command language, the command, “Bring me a slab!” may be a longer version of “Slab!”  If a dog performs a single command, fetching, with regard to an item, does the dog have any concept of the item separate from the command?  Paul Bloom, commenting on Kaminski et al. in the same issue of Science in which the article about Rico appeared, argued that when children learn a word, such as “sock,” they do not interpret is as “bring the sock,” or “go to the sock.”  Bloom said that children “appreciate that the word refers to a category, and thereby can be used to request a sock, or point out a sock, or comment on the absence of one.”  He doubted that Kaminski et al. had established that Rico understood reference. 

Pilley and Reid used three different commands, take, paw, and nose, and combined these words with three different objects.  “Take” required picking up the object with the mouth. “Paw” required touching the object with a front paw.  “Nose” required touching the object with the nose.  Chaser was 100% accurate across 14 trials, performing the correct command to the correct object.  This indicated that, to Chaser, the nouns referred to objects.  These researchers said that Chaser had demonstrated “combinatorial understanding” of two-word phrases, at least similar to the way children do. 

Daniela Ramos and Cesar Ades (2012) of the University of Sao Paulo worked with a female mongrel dog, Sofia, who obeyed two commands, “point” and “fetch,” which were combined with four nouns, “key,” “ball,” “stick,” and “bottle.”  “Bear” was an additional noun added later.  Point and fetch commands were also given as to objects (toothbrush, rubber teether, etc.) as to which Sofia had not been given a name.

After learning the names of the objects and the commands, trials involved both objected and action requests: ball ... fetch, ball ... point, key ... fetch, key ... point, bottle ... fetch, and stick ... point.  After first hearing the name of the object, as Sofia approached it the action request was added, but if Sofia approached the wrong object, the trainer blocked her and said “no.”  Sofia then returned to the initial position and the object request was repeated, followed by the action request.  The number of object requests was progressively increased.  Object choices were correctly made between 64% and 79% of the time, and command requests were between 93% and 98% correct.  As to Sofia’s higher performance for commands as opposed to selecting objects, Ramos and Ades state:

“This intriguing difference which deserves a more thorough examination, may relate to the history of dog breeding, during which many breeds were developed for herding, tracking, etc., cooperative tasks in which commands for action (not for object discrimination) prevail.”

Although not part of their current paper, Ramos and Ades mention that in additional research they combined several commands in sequence, turn right or turn left, with fetch and point.  They report that in this experiment Sofia was also highly successful, confirming her “capacity to take into account and combine information items of a different nature.” It should be noted that combined commands are commonly used with working and sport dogs.  In field trials, a dog can be given a fetch command, followed by a right or left signal given by voice, arm movements, or a whistle.  Similarly, sheep dogs will be guided during a task of moving sheep with additional commands: stop, forward, left, right, back, and others, alone or in rapid combination.

These researchers purposefully did not give two object-action requests through most of the experiment, stick fetch and bottle point.  They wanted to see if, at the end of the experiments, Sofia would perform as well to unfamiliar pairings of commands and items as she had to pairings that had been used many times.    

In the next phase, object and action requests were given as single commands, so that Sofia, while looking at the experimenter, was given a ball point command and then released to respond.  Objects were increased from two to four as the sessions proceeded. Sofia chose the correct object 85% of the time for two objects, 68% for three, and 49% for four, but action terms were obeyed correctly 90% of the time or better regardless of the number of objects.  Various control tests involved introducing novel conditions, such as the experimenter wearing sunglasses or covering her face with a cloth band, but Sofia continued to perform above chance levels. 

Sentences were now inverted, with actions preceding objects, so that Sofia would hear fetch ball, point ball, fetch key, etc.  The researchers believed that inverted requests would not lead to correct performance if a single-stimulus hypothesis explained canine behavior.  The researchers found that item reversal did not affect Sofia’s performance.  Correct responses, however, decreased from 80% at the first session, 70% at the second, and 60% at the third. (Falloffs in performance may be explained by experiment fatigue or other problems.  See Jezierski et al., 2012, on training cancer detection dogs.)

A teddy bear, a novel item, was then incorporated into Sofia’s repertoire in simultaneous object-action requests, bear fetch and bear point.  Additional objects were added so that the bear had to be selected from among other objects.  Sofia had 75% correct responses for bear fetch and 60% for bear point. 

As noted above, Sofia was not exposed during the first phases to two object-action requests, stick fetch and bottle point.  Now these were given but Sofia did not do well at all.  She had only three correct responses out of ten for stick fetch requests and 3 out of 10 for bottle point requests.  This was not a successful level of responding, which the researchers could not explain.  They note that previous training for stick point and bottle fetch might have effectively trained Sofia only to perform one action with each item.  

Pointing vs. Naming

When an adult points to an object, infants as young as 13 months expect that the word she is saying is the object’s name.  (Gliga and Csibra, 2009)  Susanne Grassmann, Juliane Kaminski, and Michael Tomasello of the Max Planck Institute in Leipzig designed some experiments to determine how dogs react to pointing when an experiment involves objects whose names they have learned.  They used two Border collies, Paddy, who knew about 60 object names, and Betsy, who knew 300.  The researchers lined objects in a row and first asked the dogs to fetch familiar items.  They then pointed to a new object and gave it a name while asking the dog to fetch it.  “Fetch the blicket,” or some other nonsense term.  The dogs relied on pointing when they had not been taught the name used before. 

But what would happen if the verbal command to fetch was in conflict with the name the dog had been taught?  The researchers found that the dogs first approached the object the experimenter was pointing at, but fetched the object whose name they knew.  The researchers believed that pointing is helpful to dogs in finding an object whose name they know, and that they take the pointing as a “spatial cue that leads them to walk in the indicated direction.”  While spoken labels “are directly mapped to individual objects through the dogs’ word training," “dogs may interpret pointing as some kind of directive ordering them where to go instead of informing them about/referring to objects in the vicinity.” 

Grassman et al. noted that prior research had indicated that for humans, pointing and words are both referential, i.e., human infants interpret the object pointed to as being given the name spoken simultaneously.  Children even rely on pointing when pointing and naming conflict.  For dogs, the pointing may tell them where to begin the search, but may not be interpreted as an attempt to name the object pointed at.  The researchers, referring to the study by Pilley and Reid, note that pointing was used in training Chaser the names of objects, and acknowledge that “Chaser’s experience with pointing and naming might be more similar to children’s experience.” 

Do Dogs Understand Categories of Objects? 

Pilley and Reid sought to determine if Chaser could understand that a noun referred to a category of objects, whether, in other words, she could understand a common noun as opposed to a proper noun.  They therefore designed an experiment to determine if Chaser could correctly indicate that her toys belonged in three categories, “toy,” “ball,” and “frisbee.”  Balls and frisbees (disk-like objects) were categories within toys.  Toys were objects that Chaser was allowed to play with.  Non-toys were objects around the household that were on shelves, desks, tables, and in closets, and Chaser was not allowed to play with these items. 

For training, eight toys were placed among eight non-toys in a room.  Chaser was told to fetch another toy.  If what she brought out of the room was not a toy, she was told, “No, that is not a toy.”  If she retrieved correctly, she was reinforced.  Training continued until Chaser only retrieved toys.  The researchers argued that the fact that Chaser was soon able to retrieve eight toys correctly, and leave eight non-toys untouched, indicted that she comprehended the common noun “toy” as a label for the toy category, even though she also knew each object by its proper-noun name.  She was also perfect in selecting balls and frisbees. The researchers state:

“Thus, Chaser learned that names of objects may represent categories with many exemplars – for each of the three common nouns, she mapped one label onto many objects. Membership in two of the categories, ‘ball’ and ‘Frisbee’, could be discriminated based on common physical properties [shape, in particular]. For example, all balls had similar round shapes, and all Frisbees were shaped like a disk. The toy category required a more abstract discrimination. The ‘toy’ and “non-toy” objects did not differ in terms of common physical characteristics. Thus, discrimination of whether an object is a toy or not was unlikely to be based upon identifying physical ‘toy-like’ features. However, Chaser learned that she could play with toys, but she could not play with non-toys – they differed in functionality. While this study was not designed to assess how Chaser created categories, it seems likely that the common noun ‘toy’ reflected a more abstract type of categorization than did ‘ball’ or ‘Frisbee.’” 

It is perhaps to be noted that balls and frisbees do not differ only in shape.  Their movement characteristics are somewhat different.  Balls bounce and roll, which makes them easy to chase, but frisbees are more easily caught in mid-air. The two types of toys are also held differently in a dog’s mouth.  In any case, Chaser understood that an object may have more than one name, and in some cases three.  Thus, an item could have a proper name, but also be a ball and a toy.

This is an interesting study and needs to be broadened.  Can dogs form categories of things that they cannot pick up but only go to?  If Chaser were told to select a ball in the first part of a trial, then a frisbee, how successful would she be if told to fetch a toy when the only objects were balls and frisbees? If a dog were told to get an object from Room A, while the same object was in Room B, and in a field outside the rooms, could the dog learn to recognize a specific room but tell rooms from the field?  

When Learning to Identify an Object, What Features Are Most Important to the Dog?

Griebel and Oller, the Yorkshire terrier who had been taught the names of 120 objects, noticed that Bailey, upon first encountering a new item, could act in several ways. They describe the difference between two new toys, Triceratops and Dora the Explorer:

“Bailey’s relationship with Triceratops seemed to be different from that with Dora the Explorer, because she would often refuse to give Triceratops up after a retrieval and would carry the toy around as well as shake it as if in a predator-prey interaction. She never did these things with Dora the Explorer, which she simply dropped in front of her owner after each retrieval. This difference in behavior resulted in longer periods of initial informal training for Triceratops than for Dora the Explorer….”

I have noticed that in playing with stuffed toys of squirrels, rabbits, and a large frog, my dog sometimes shakes in this predator-prey manner.  Is this a visual recognition of potential prey, since the toy would presumably not smell any different from other stuffed objects?  Most of the research on name recognition in dogs has not considered this kind of qualitative difference in learning names.  As a dog like Chaser learns more and more names, does this learning take on a routine nature where such variation disappears? 

Three scientists at the University of Lincoln in the UK, Emile van der Zee, Helen Zulch, and Daniel Mills (Zee et al., 2012), asked whether dogs learn names of objects similar to the way children do.  Earlier research (Landau et al., 2008) had concluded that when young children learn to connect a name with a new object, they generalize the meaning of the name to objects that are similar in shape, but not objects that are similar in size.  Thus, when taught that a specific U-shaped object is called a DAX—a name made up for purposes of the experiment—they will identify other U-shaped objects as DAXES.  They do not identify objects of the same size or texture, but not the same shape, as DAXES. This team, working with a male Border collie named Gable, concluded that dogs initially generalize to size, but after becoming more familiar with an object, texture becomes more important.  Gable had learned the names of at least 43 different objects from his owner. 

In the experiment, Gable was taught new objects and names by holding the new object in view and saying its name several times, after which the dog was allowed to play with the object while the name was repeated again.  He was then asked to fetch the object.  Next, the object was placed among five to 15 other objects for which Gable knew the names and Gable was told to fetch the new object from the set.  Objects used in the experiments were made from foam swimming boards, sometimes covered with different cloths to provide different textures. 

After being taught that a DAX was a U-shaped object of a certain size (over a ten-minute period), Gable was then told to “get dax” and given a choice between two objects, one of which was the size of the object on which he had been taught, but not its shape, and one of which was U-shaped, but larger.  The researchers found: “Gable linked the word dax to DAX-sized objects in ten out of ten cases in which he was given the choice between a DAX-sized object and a larger object …, thus confirming that Gable generalized the meaning of the word dax to other objects that were of the same size as the DAX object, irrespective of their shape or texture.”  During this phase of the experiment, Gable did not demonstrate a preference for objects that had the same texture as the standard DAX object. 

Four months after these experiments, Gable took the DAX home for 39 days and was taught to link the word “dax” with the object on a daily basis.  Now, however, when the earlier DAX experiment was repeated, Gable “tended to generalize his word knowledge to objects of the same texture … but not objects of the same size.”

The researchers asked:

“Why do object names refer to object shape for humans and to object size or even texture for Gable? In order to answer this question it is necessary to establish that objects can only be named if they are categorized or identified. Although touch can give humans reliable information about object identity, we primarily rely on vision for object identification. Visual object shape is the most important feature which makes categorization or identification of solid objects possible, and our cognitive system therefore relies mostly on shape for object naming. The evolutionary history of our sensory systems – with vision taking priority over other sensory systems – seems to have primed humans to take into account visual object shape in object naming tasks.”  

When we ask a child, "Show me what a frisbee looks like," the child may take a crayon and draw a circular object.  The child will replicate what comes to mind.  It appears that with a dog, the dog will think of texture first if it knows the object well, then of size, and may never imagine the object's shape at all.  Although the dog's smell was on each object it learned, this does not mean that the objects did not have separate smells which might also have figured in how the dog selected an item to obey a particular command. And how do we explain Bailey's shaking of Triceratops?  Is it possible that in learning names, in identifying objects with human words, dogs use behaviors and neural pathways that are distinguishable from how they perform functions such as hunting or killing prey once it is apprehended? 

The results of van der Zee et al. may explain something about why Grassman et al. found that the dogs follow pointing gestures in looking for an object they have been verbally directed to find.  Since they are not looking for the shape of the object, they have to get close enough to determine texture, size, and probably smell.  The shape is of less significance, so that pointing cannot be ignored until they identify other features.  

Perhaps sophisticated brain neurology may be necessary to specify the order of attributes that mean most to a dog in creating such references as are used by the animal. 

Exclusion Training and the Question of Fast Mapping

One of the most dramatic conclusions reached by Kaminski et al. in 2004 was that their results indicated that dogs, like children, could learn the meaning of a new word after a single or very few exposures, a process called “fast mapping.”  The “exclusion experiment” that led them to this conclusion was described by the researchers as follows:

“To assess Rico’s ability to fast map, we placed a novel item together with seven familiar items in an adjacent room…. In this so-called identification task, we conducted a total of 10 sessions in which we introduced 10 novel items. In the first trial of a session, the owner always asked Rico to bring a familiar item, and in the second or third trial asked him to bring an item using the novel name…. Rico retrieved the novel item from the first session on and was overall correct in 7 out of 10 sessions…. Apparently, he was able to link the novel word to the novel item based on exclusion learning, either because he knew that the familiar items already had names or because they were not novel.”

Four weeks later Rico correctly retrieved a target item he had not seen since the original experiment in three out of six sessions, a retrieval rate “comparable to the performance of 3-year-old toddlers.” 

Pilley and Reid provided some support for Kaminski et al., finding that Chaser could select a novel item whose name she had just learned from a group of object that included three novel and four familiar objects.  They noted, however, that in exclusion choice trials, retention was reduced when Chaser was tested after ten minutes and was “essentially non-existent” after a 24 hour delay. They observed that “additional pairings with the name (playing with the object) were necessary for retention intervals of an hour or more.”  Although these researchers referred to mapping, they avoided a claim for fast mapping. 

Griebel and Oller doubted that the research with either Rico or Chaser demonstrated fast mapping.  They suggested instead that Rico and Chaser, and even children in similar experiments, may have succeeded by what they call “extended exclusion,” with no fast mapping at all.  That is, the dogs may have succeeded in retrials because they retained the names of items they had been taught as well as the one item whose name they had heard once but for which they had been rewarded, excluding the objects whose names they had never heard and for which they had received no reward.  If so, the novel name might not be a referent in the way Rico’s and Chaser’s researchers had argued. 

These researchers attempted to use exclusion tests with Bailey, but found when asked to retrieve a new toy from a group that included seven known toys, she was not successful above chance.  She clearly did not like this type of task:

“During these trials, Bailey showed signs of agitation, barked often, and refused to go to retrieve the requested new item several times, and this occurred selectively on the trials with the new items. Reviewing the video of these sessions, we saw that in two cases Bailey handled the new toy for a while and even carried it a few steps before dropping it and picking up one of her known toys to bring to her owner.”

Griebel and Oller argued that the research on Rico and Chaser had been “over-interpreted as indicating fast mapping.”  They state:

“There clearly remains the logical possibility that success on the tasks can be based on exclusion alone (choose the novel item) plus an extended kind of exclusion (choose the relatively novel item, the one that has been recently seen or rewarded).”

Issues for Further Research 

Some of the papers described mention that Border collies may be exceptional in terms of their abilities to learn names of objects.  They have been bred for identification and separation of objects from groups of often similar objects.(L.E. Papet agrees with the high assessment of the intelligence of Border collies, but notes that "the pitfall with the breed is that it is not uncommon that intelligence of the dog allows it to outsmart the handler.")   It is to be hoped that more types of dogs, and more dogs with other specialized skills, such as pointers and retrievers, bloodhounds and beagles, will be tested for their name-recognition and other specific abilities as well.  

For instance, what does a bloodhound do when it reaches the end of a track?  How does it know that it has found a match?  A forensic scientist might say that the size and tread pattern of the shoes of the individual the dog led to match shoeprints at the crime scene.  The dog, however, may be satisfied at the end of the track that the objects found on the individual's feet, shoes made of rubber and leather and sufficiently porous for a certain amount of seat to seep through, must be what was being followed and are all that will be found. At least no stronger odor seems to be available.  Is size significant as well in the dog's decision not to search further?  Shape is not as relevant as it is to the forensic scientist, but is it irrelevant?

Does a pointer stop and point because of the bird's shape and smell?  How much does the quarry's size figure into the signal of recognition?  Does such a dog have a learned sense of movement?  Movement in the experiments described here was part of the dog's training when the owner played with a new toy.  When the owner told the dog to fetch or point, however, the object was idle.  Does a sighthound identify shape better than other breeds?  Did Bailey, the Yorkshire terrier, shake Triceratops because he shares in such a characteristic? Or does a sighthound have a heightened sense of movement when it pursues the hare, a unique understanding of trajectory that is only vestigial in other dogs? Learning to find objects that have been given verbal identifiers by people is an aspect of interspecies communication, but hunting need not be.  Have the researchers in these studies always considered the ramifications of this difference in their interpretations? 

Questions can be imagined regarding identification patterns in any functional breed, whether there are heightened sensory skills and parts of the brain more developed than in other breeds.  Neurology will eventually have to contribute to answers where experiments can identify behavioral differences (hopefully avoiding some of the horrors of earlier brain research). Of course we also have to be concerned that as breeding becomes dominated by show dog objectives, behavioral and neurological differences correlated with ancient functions are actually disappearing, as Kenth Svartberg has demonstrated.   

Conclusion

Dogs can be taught to retrieve a great many objects by name.  They can be taught to perform a number of different actions with respect to an object the name of which they have learned.  Whether they can do so quickly, in a manner that demonstrates fast mapping, must be considered as open to doubt at the moment. It appears they learn names of objects by first identifying the object’s size and later its texture. 

Could not the different smells of objects have been useful in distinguishing objects in these experiments?  In all the studies described here various controls were used to assure that the dogs’ own scent was present on each item or otherwise eliminated as an identifying factor, but many of the objects undoubtedly contained materials that produced odors.  Given the number of objects involved, however, the experimenter's human sensory limitations prevented considering how smell operated.  Besides, the general purpose was to determine if dogs could identify objects with words, not so much how they did it. In the one study that focused on the identification process, the researchers sought to avoid scent cues by covering objects with the same types of materials, two layers of cloth, and the materials were stored together.  Assuming this was effective in eliminating scent as a possible identifier for the dog in that set of experiments, it does not eliminate the likelihood that dogs would probably rely heavily on scent were this factor not controlled. (Just consider the research on volatile organic compounds in cadaver dog and cancer sniffing work.)

It also appears that dogs may be able to learn some categories, toys, balls, frisbees, but this may depend on the objects fitting into a category being within a certain size range, or it may require that the dog learn to play with items in the same category in the same way (balls that bounce, frisbees that float in the air).  Additional research will be needed to more precisely delineate the categorization skills dogs have.

The Shepherd's Dog (Edwards, Cynographia Britannica, 1800)
So what can we say about the sheep thief?  Is the account possible?

It seems unlikely that merely pointing from outside a fold would have supplied the thief's dog with enough information to come back later to separate and herd away the indicated animals. Even if the thief pointed as he walked, saying “A … B … C” and so on, moving his arm each time, it seems beyond the ability of dogs that the indicated sheep would be remembered hours later. It would only be possible if the thief’s dog, like Chaser, had been taught to use pointing as more than a helpful direction in which to go and look for an item, but identification would seem to require a more extended process. Even if that were true the dog would have needed exceptional skills.  If the thief touched the sheep, particularly if the thief had grease or some other pungent substance on his hand, the transferred scent becomes possible as an identifier. 

It is easier to postulate that there was some categorization of the sheep to be stolen, such as younger ewes, which had been part of the combined criminal enterprise such that the dog knew the master’s wishes.  Either that or the dog understood that his task was to take as many sheep from the flock as he could drive quickly to the rendezvous point, perhaps a road at the end of a stretch of common land where flocks of various owners grazed.  Presumably the dog followed his own and his master’s tracks back to the location of the particular flock, one without guard dogs living with the sheep, separated out a dozen or so, then herded them to his master.  All of these tasks are well within the repertoire of a sheep dog.

Thanks to Eric Krieger, Richard Hawkins, L.E. Papet, and Kingsbury Parker for comments and suggestions.

Sources:
  1. Ash, Edward C. (1927). Dogs: Their History and Development.  Ernest Benn, Ltd., London.
  2. Aust, U., Range, F., Steurer, M., and Huber, L. (2008). Inferential Reasoning by Exclusion in Pigeons, Dogs, and Humans.  Animal Cognition, 11, 587–597.  
  3. Bloom, P. (2004). Can a Dog Learn a Word?  Science, 304(5677), 1605-1606.
  4. Derjean, D., Moussaddy, A., Atallah, E., St-Pierre, M., Auclair, F., Chang, S., Ren, X., Zielinski, B., and Dubuc (2010). A Novel Neural Substrate for the Transformation of Olfactory Inputs into Motor Output.  PLOS Biology, 8(12), e1000567 (discussing "the neural basis underlying motor responses elicited by olfactory inputs in higher vertebrates," using lampreys, but referring to prior research on tracking behavior in dogs).
  5. Erdöhegyi, A., Topál, J., Virányi, Z., and Miklósi, A. (2007). Dog-Logic: Inferential Reasoning in a Two-Wy Choice Task and Its Restricted Use. Animal Behavior, 74, 725–737.
  6. Gliga, T., and Csibra, G. (2009) One-Year-Old Infants Appreciate the Referential Nature of Deictic Gestures and Words.  Psychological Science, 20, 347-353.
  7. Grassmann, S., Kaminski, J., and Tomasello, M. (2012). How Two Word-Trained Dogs Integrate Pointing and Naming.  Animal Cognition, 15(4), 657-665.
  8. Griebel, U., and Oller, D. (2012). Vocabulary Learning in a Yorkshire Terrier: Slow Mapping of Spoken Words. PLoS ONE, 7(2): e30182. (doi:10.1371/journal.-pone.0030182).
  9. Kaminski, J., Call, J., and Fischer, J. (2004). Word Learning in the Domestic Dog: Evidence for “Fast Mapping.” Science, 304 (5677), 1682–1683.
  10. Landau, B., Smith, L.B., and Jones, S.S. (1988). The Importance of Shape in Early Lexical Learning.  Cognitive Development, 3, 299–321.
  11. Pilley, J.W., and Reid, A.K. (2011). Border Collie Comprehends Object Names as Verbal. Behavioural Processes, 86, 184–195.
  12. Ouattara, K., Lemasson, A., Zuberbuhler, K. (2009). Campbell’s monkeys concatenate vocalizations into context-specific call sequences. PNAS 106(51), 22026–22031.
  13. Ramos, D., and Ades, C. (2012). Two-Item Sentence Comprehension by a Dog (Canis familiaris). PLoS ONE, 7(2): e29689. (doi:10.1371/journal.pone.0029689).
  14. Smith, C.H. (1840). Dogs: Canidae or Genus Canis of Authors.  W.H. Lizars, Edinburgh.
  15. 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.0049382.
  16. Wittgenstein, L. (1953). Philosophische Untersuchungen/Philosophical Investigations.  Basil Blackwell.  The fourth edition (2009) has been published by Wiley-Blackwell after John Wiley and Sons’ acquisition of Blackwell in 2007.  Sections 2 and 19 particularly discussed here.

No comments:

Post a Comment