Border collie syntax?
« previous post | next post »
I have been unabashedly contemptuous of previous stories about dogs learning to understand human languages (for example, I hammered home the point that fetching named objects is not understanding language in this 2004 post, which Mark Liberman followed up here). I think it is incumbent on me to acknowledge further developments in the area that actually lead to refereed publications, so let me point out that Learning and Motivation has now published a paper about some experiments purporting to show that a 9-year-old border collie called Chaser has learned some rudimentary syntax. For example, Chaser can differentiate To ball take Frisbee from its inverse, To Frisbee take ball, and perform the right action in each case. This Science News article gives a fairly full account of the results.
Me? I honestly don't know what I think about this yet. Maybe you want to comment below? I'm cool with that. Remember, though, that on the Internet nobody knows you're a dog, so if you are a dog, please admit it; it's a matter of declaring your interest in the controversy.
Dylan said,
May 24, 2013 @ 2:14 pm
This still sounds pretty far-fetched.
MonkeyBoy said,
May 24, 2013 @ 2:43 pm
Given that dogs differ from wolves in that they are particularly good at picking up human intentions I wonder how they controlled for "Clever Hans" effects.
I would like to see the experiments run with the dog in a room with no people and the commands coming through a speaker.
X said,
May 24, 2013 @ 2:58 pm
According to the article, they controlled for Clever Hans by using a student who did not know the names of the toys. Each toy was given a proper name (e.g. "Tobey", "Spike"), so the student probably could not figure them out to introduce bias.
Theodore said,
May 24, 2013 @ 3:17 pm
And I, for one, welcome our new border collie overlords.
MattF said,
May 24, 2013 @ 3:24 pm
We perceive the order of words in a phrase as 'syntax', but is that true for a dog? Chaser's a very smart dog, so it may well be capable of distinguishing 'blugleforfax' from 'forfaxblugle'– but that's a long way from syntax.
X said,
May 24, 2013 @ 3:48 pm
@MattF: They controlled for that by using words that the dog knew outside the context of the test sentences. Although the dog was trained on other sets sentences with meaningful ordering, the dog was not trained with the specific toys used in the experiment. The dog had to mentally connect the meaningful ordering with the separately learned meaning of the words to reconstruct the desired action. This requires syntax.
Jon Weinberg said,
May 24, 2013 @ 3:49 pm
@MattF: but the writeup says that Chaser got commands right even when they included object names that hadn't been used in sentence training. That's the impressive part.
Rod Johnson said,
May 24, 2013 @ 5:22 pm
Why is that a long way from syntax? Words in a rule-governed syntagmatic relationship—that's syntax.
BW said,
May 24, 2013 @ 5:56 pm
In Experiment 3, not much syntax processing was required: "The two direct objects were always placed in the bedroom and the two prepositional objects were always placed in the living room."
Since the objects were unique, it would not matter to the dog whether I say "take A to B" or "to B take A" – if only A is available in the bedroom, the dog knows to pick up A and run to the living room with it. Of course this is still a difficult task because of the memory demands, but I think that this experiment does not test syntax.
In the other experiments, it seems to me that there could still have been regularities in taking items from the left side to the right side or vice versa. The paper describes that the sides were alternated in some regular pattern and not (pseudo-)randomized.
I also think that they did not really control for Clever Hans because the dog's owner (=trainer) was always in the room. Without seeing the video, it seems possible that the dog picked up changes in breathing or posture (or some other subtle signal that we would not even perceive) when it ran towards the correct/incorrect side or moved its mouth towards the right/wrong item. They checked whether the dog looked at the owner, but it would not have had to look directly if it could hear him shifting weight or changes in breathing.
I still think it's impressive, but perhaps not quite rigorous enough to be really convincing.
John Swindle said,
May 24, 2013 @ 7:46 pm
Humans and dogs have interacted for a long time, and we had to get syntax somewhere. It's clear that we didn't get it from cats.
Jeff DeMarco said,
May 24, 2013 @ 9:17 pm
John Swindle said: "It's clear that we didn't get it from cats."
Q.E.D.
Mark F. said,
May 24, 2013 @ 10:27 pm
I think it's a mistake to treat syntax as this magical and unitary thing that was given to humans by Prometheus along with fire. We may well have brains that are optimized for syntax, and be the only animals that do, but that doesn't make it impossible for some amount of syntax to be learned by less specialized faculties. I think of this result (to the extent that I believe it) more as a demonstration of a remarkable ability to generalize than as a breaking of some language barrier.
I am still confident that we are the only species that has discussions about whether we are the only species that can have discussions.
Sollers said,
May 25, 2013 @ 3:02 am
I have no real problem with this – given that it's a Border Collie. The thing that I have always found astounding about their use as sheepdogs is that they can tell left from right; this is a huge conceptual thing (and one that my very intelligent daughter still can't manage consistently in her 30s).
Breaking the task down: the dog needs to understand two nouns (carefully controlled trials have shown some dogs have a very large vocabulary), an understanding of verbal content (even more dogs can handle this, every time they perform an action such as "sit" or "fetch"), and an ability to understand combinations of these (the dogs mentioned above were able to fetch a specified toy). These correspond to what at any rate used to be called "pivot constructions" so the only question here is the last stages. One is performing an action on one item in relation to another, and the other is reacting to word order to indicate which goes to where.
On the personal, subjective level, the feeling I get with my dogs is that I'm speaking in alphabetic script while they are hearing in ideograms. I work from home, in the front bedroom, and sleep in the back bedroom, and my younger dog and I have developed rituals; if I say "Time to go to work" she runs upstairs and into the front bedroom, but if I recite "To bed, to bed, said Sleepyhead" she runs upstairs and into the back bedroom – she clearly associates these deliberately long utterances with the appropriate room. Note that time of day can vary widely, as sometimes (but not always) I go up for a rest during my lunch break. Note also that I am not giving her instructions, but expressing my own intentions as to where I am going to go, and it's up to her whether she accompanies me or not – but she's part Papillon, hence very clingy, so wants to be wherever I am; to the extent of going ahead of me when she knows where I'm going.
Ray Dillinger said,
May 25, 2013 @ 4:22 am
The guy who used to work on my car had a dog that he had trained to fetch tools by name. Ie, he'd say "get the wrenchroll" or "get the screwdriver" or "get the crescent wrench" or whatever, and the dog would find the appropriate object and take it to him. With a pretty huge vocabulary, too; this English sheepdog knew every tool in the shop down to oil filter wrenches, spark plug sockets, and O-ring compressors.
The same dog also took the command "put it away" where the mechanic would hand a tool back to the dog and it would take it to the workbench or the toolbox or wherever it belonged, so it clearly had associated each object with a location.
This falls short of syntax, but it's impressive, and I don't think this dog was getting much in the way of 'clever hans' clues from his owner — the guy was usually focused, not on the dog, but on what he was working on. He didn't even watch the dog to see what it was doing. Often nothing of him was visible but his feet sticking out from under a vehicle.
I don't find it particularly unbelievable that a dog, especially a "working breed" like a border collie, could understand some simple syntax; they've been trained and bred for millenia to respond to spoken commands. I'd still be properly astonished to find a dog using syntax in the production of new utterances.
Ralph Hickok said,
May 25, 2013 @ 10:23 am
@Ray Dillinger
" I'd still be properly astonished to find a dog using syntax in the production of new utterances."
Do you mean, for example, a dog saying "Wow-Bow" instead of "Bow-Wow"?
Ray Dillinger said,
May 25, 2013 @ 12:39 pm
@Ralph Hickok
Only if it became clear that the dog meant something different by it.
I recall a Gary Larson cartoon in which someone had developed a device to translate dog's "speech" to English, and it turned out that the dogs were just saying, "Hey!" over and over.
I think that's a reasonably accurate account of dogs barking. I mean, they pronounce it differently when they mean 'get the hell out of my yard' versus 'glad to see you,' but humans can have whole conversations by saying "Hey" with different intonations too.
Nelida K. said,
May 25, 2013 @ 7:38 pm
@Sollers @Ray Dillinger: I really can't say about syntax, but that dogs do understand what we say to them, is quite unquestionable. I had a cocker spaniel, and if I asked him (while in my den, sitting in front of the PC) "Do you want to eat?" he turned and ran like the wind to the kitchen. If I asked him instead "Do you want your yummy food?" (all this in Spanish, of course) he did exactly the same. (Note that these utterances were issued as questions, not orders). If instead I said: "I am going to do a little errand" he went and made himself comfortable on the sofa, because he knew that meant I was going out without him. But once I forgot and told him "Let's go out to do a little errand" – he stood looking at me with a "what-do-you-mean-let's-go" look in his face, so I realized my faux pas and changed it to "let's go out to the street", then he immediately went to the front door. And as to understanding intentions, when I finished working for the day and shut down my PC, as soon as he heard the shut-down music, he stood up and went to the kitchen. Without any need for me to tell him anything. So maybe not syntax, but they are able to understand language, and I don't think that we should rule out the syntax part entirely. IMHO.
Daniel said,
May 25, 2013 @ 11:27 pm
Pilley's Youtube channel seems to have the actual videos for at least experiments 1 and 2.
Chandra said,
June 3, 2013 @ 1:52 pm
@Dylan – Your comment, however, is on the ball.