Monday Pets: How Do Dogs Learn New Words?

...and what can word-learning in dogs teach us about the evolution of language in humans?

ResearchBlogging.orgWhat is involved in the learning of a single new word? Consider the word "tiger", being learned by a child with already a modest vocabulary, at least for animal words. First the child must make a new entry in the mental lexicon - that "tiger" is a word in the first place. He has to categorize it as a noun. It has to be categorized under "animal" (a supernym) and related to its hyponyms, like "Sumatran tiger." Then, of course, the child has to learn what actual *thing* the word "tiger" refers to. Now, various conceptual categories likely have to be restructured. Before, the child might have referred to tigers as "cats," but now the child has must conceptually distinguish cats from tigers. Sometimes, the child has to accomplish all of this without explicit instruction; he or she may be exposed to a word casually, or in the course of conversation. Early research showed that children indeed were able to learn new words after just a single casual exposure. As you can see, learning only one new word involves learning a considerable amount of new information.

The process by which a child learns a new word after only one exposure is called "fast mapping." And kids "fast map", well, fast. And often. From 2 years of age, typical English-speaking children add about ten new words a day to their vocabulary until they reach an average vocabulary size of 60,000 words by high school graduation.

Is the ability to fast map unique to language learning, or does it reflect more general cognitive learning skills that may be shared with other animals? Meet Rico.

i-e2b9b3e4ad589a95b1390ef88136d993-rico in flowers.jpg

FIgure 1: Rico, a border collie.

Rico, a border collie, was born in December 1994. At the time of this research, he was about 10 years old. He was first introduced to the fetching of items at 10 months of age, when his owners would place various items around the room and ask the dog to retrieve one of them. He was rewarded with food or attention if he retrieved the correct object. Over the years, he was introduced to more and more items, and by the time of this research, his owners reported that he knew the labels of over 200 individual items, mostly toys and balls, which he could correctly retrieve upon request. It was unclear, however, whether Rico was simply responding to subtle behavioral cues of his owners, as opposed to actually knowing the words for the objects. Was Rico a modern-day canine version of Clever Hans?

Experiment 1 was designed to assess Rico's ability to retrieve various items under precisely controlled conditions, to ensure that his owners weren't unintentionally cuing him to the proper answer. The owner waited with Rico in one room, while the experimenter arranged a set of eight of Rico's items in a second room, before joining the owner and dog. Then, the owner asked the dog to bring two randomly chosen items (from the set of possible items set out) from the next room. This ensured that while Rico searched for the proper item, he could not see the owner or the experimenter. He retrieved a total of 37 out of 40 items correctly, indicating that that he indeed knew the labels of those objects. Amazingly, Rico's vocabulary was larger than any other dog studied, and was comparable to the vocabularies of language-trained monkeys, dolphins, sea lions, and parrots.

Experiment 2 addressed Rico's potential ability to fast map. The setup was the same, but this time one new item that Rico had never seen before (and for which he did not know the name) was included together with seven familiar items. In each of 10 sessions, a different novel item was introduced each time, for a total of 10 new items. There were 8 trials per session (one trial per item, per set), for a total of 80 trials. Of course Rico took breaks between each session.

Video 1: The first three trials from the first session, from the supplementary materials available here.

In this video, Rico is first instructed to bring two familiar items: first the "tyrex" (the blue dinosaur) and next the "weihnachtsmann" (the little red doll). Subsequently, a novel word ("sirikid") is used to ask for the novel item, the white bunny. Perhaps now it would make sense to mention that Rico and his owners lived in Germany. Rico successfully retrieved the correct new item (the white bunny) in the first session, and continued to do so in 7 out of 10 sessions for additional new items. Amazing! The experimenters reasoned that he was able to link the novel word with the unfamiliar item based on exclusion learning - because he already knew the familiar items names' or because they were not unfamiliar.

Experiment 3 was conducted four weeks later, and was designed to assess Rico's retention of the relationship between the novel words and the novel items. Three of the novel items from experiment 2 were used again in this experiment. Again, the procedure was the same as in experiments 1 and 2. The target item, along with four additional novel items, and four familiar items were laid out in the room, for a total of nine items. As before, Rico was first asked to bring one or two familiar items before being asked for the target item. Rico correctly retrieved the target item in three out of six sessions. It is worth noting that in the failed trials, Rico never brought a familiar item when asked for the target item; he always brought one of the other unfamiliar items. Fifty percent success doesn't seem like much, but consider that his success rate was comparable to three year old humans, in similar tasks.

Overall, these experiments indicate that Rico consistently learns to associate arbitrary acoustic patterns (i.e. human words) with specific objects. His extensive experience with acquiring object names has allowed him, apparently, to learn that objects have names. Further, he has been able to learn the association between a new word and a new object based on exclusion. He is also able to store this association in memory, since he was able to correctly retrieve the target item from a set of familiar and unfamiliar items both immediately as well as one month later.

It may be the case that Rico's amazing word-learning ability is the result of an "exceptional mind" or his extensive experience with learning word-object relations. Even if that is the case, as for any other cognitive capability, Rico's performance can be broken into its constituent building blocks:
(1) the principle that objects have labels,
(2) a general learning mechanism, driven by the exclusion principle, and
(3) the ability to store that knowledge in long-term memory.

Young children, of course, have a much broader understanding about words and their functions than Rico does. Children (even babies) can readily distinguish among nouns and verbs, for example, based on their acoustic structure. Children are also able to use newly-learned words productively in speech, while dogs, obviously, do not. However, this research suggests that the ability to attach meaning to specific sounds may have emerged earlier than, and evolved independently from, the ability to flexibly produce those sounds. Some of the perceptual and cognitive building blocks that underlie speech comprehension may have been in place before humans began to talk. What appears to be a complex linguistic skill used by human children to learn new words may depend on simpler cognitive building blocks also present in at least one other species, the domesticated dog.

Kaminski, J., Call, J., & Fischer, J. (2004). Word Learning in a Domestic Dog: Evidence for "Fast Mapping" Science, 304 (5677), 1682-1683. DOI: 10.1126/science.1097859

More like this

Toddlers learn new words at an astonishing rate—an average, according to Steven Pinker, of over a word every two hours. Yet attempts to drill children to improve vocabulary are often frustrating. Kids seem to learn words better through observing the environment than they do by rote. So what exactly…
The Sapir-Whorf Hypothesis -- stated in its strongest form -- claims that language determines thoughts: if a language doesn't have a means of expressing a particular idea, then people speaking that language can't even conceive of that idea. This strong form has long since been rejected: There are…
Have you noticed how you tend to remember things better after you've figured them out for yourself rather than listening to someone else's explanation? Well, this phenomenon is typical for toddlers, too. According to a study published today, toddlers have an easier time learning new words when…
Earlier this week I wrote about the developmental and evolutionary origins of large number representation. A series of studies in human infants, monkeys, rats, and fish demonstrated that animals and humans spontaneously represent large (>4), abstract, approximate numerosities. Animals, human…

Great study! How language is learned is especially interesting to me right now, because I have a child that I care for who has a greatly reduced vocabulary. He is 3YO and has about 40 words in his vocabulary. But after reading this study, I'm wondering how much the building blocks of vocabulary change when a child is raised in a bilingual home. Do you know of any studies on that?
Thanks ~G

I love this post. Have known about Rico somewhat indirectly for a while - merely because I am somewhat partial to herding breeds and their *obviously* superior intelligence. Thanks for writing this up and giving a little personal background on the dog, and experimental background on his learning paradigms. So cool!

This kind of conclusion based on this kind of evidence always bothers me. How do we know for sure that Rico understood that the words were *symbols* that represent physical objects, as opposed to merely being sounds that *signalled* to him that he could earn a reward if he retrieved a particular object?

How is this conclusion different from saying that monkeys who are trained to press a button when a screen flashes red (as opposed to any other color) have learned that "red" is the symbol in human language for "press"?

By Tim Martin (not verified) on 10 May 2010 #permalink

@3:

I wrote:

these experiments indicate that Rico consistently learns to associate arbitrary acoustic patterns (i.e. human words) with specific objects

I'm not making (and neither are the investigators) any grand statement about the dog, beyond this. He associates arbitrary acoustic streams with certain objects. But, then again, how is this any different from human language comprehension?

Its easy to think of language as non-arbitrary, but human language *is* a series of arbitrary sounds that we've assigned meaning.

And the main difference between Rico and the language-trained monkeys is that he learns and remembers new words spontaneously, without explicit instruction, as evidenced by experiments 2 and 3. It takes most primates or dolphins weeks, months, or even years to amass the vocabularies that they have. Rico learned new words after a single exposure, and remembered 50% of them one month later, which is on par with 3 year old children. That's what's so amazing.

I don't disagree with you on your last paragraph; Rico's memory and ability to learn without explicit instruction are laudable.

The "grand statement" of yours that I'm questioning is that Rico was learning *words* at all (where "learning a word" entails understanding that the word is a label, or symbol as I like to call it, that stands for something else).

Your position seems to be that associating an arbitrary acoustic pattern with a specific object demonstrates a linguistic ability, and demonstrates that an animal has learned that that acoustic pattern is the name for the object. If so, why do the following not demonstrate linguistic abilities?

1. Dogs associating the ringing of a bell with receiving food.
2. My dog associating the sight or sound of me picking up his leash with us going for a walk.
3. Monkeys associating an image on a screen (because symbols don't have to be only acoustic) with the opportunity to press a button for a reward.

Why do we not say that 1. a ringing sound is the word for "dinnertime!", 2. the sound of a leash clanking against a counter is the word for "a walk", or 3. a red image is the command for "Press"?

If you're like me, you don't consider the above scenarios to be proof of any kind of linguistic ability. You probably also don't consider these stimuli to be words. But how are the above any different from a dog associating a sound ejected from the mouth of a human with the knowledge that if the dog fetches a specific object, it will be rewarded?

My contention is that in all of these cases, the relevant stimuli are *signals* that impart information to the creatures involved. But they are not symbols, which would *stand for* certain objects or concepts.

Mark Seidenberg's letter to Mark Liberman on this topic gives more information, as does Seidenberg's paper referenced in that link. He also argues in the paper that there is reason to doubt that even primates understand the words they learn *as* words (i.e. symbols), and those same arguments apply to Rico.

By Tim Martin (not verified) on 10 May 2010 #permalink

Super awesome study Jason! I remember in HS I babysat a boy from 4 weeks old till 3 years old... once when he was two I told him to get his green cup he had earlier that was in the other side of the house, assuming he would just give me a blank stare. But instead he went all the way to the kitchen and brought back the exact cup! It was great to see that developmental connection.... Now if only I could get Lucy to do the same...I think I would have faint in disbelief!

@5
The signals you discuss do not differ much from language in the sense that these signals indeed seem to function as symbols to indicate certain events or things; they activate a mental presentation in some form or another. For instance; if I say the word 'ball' to my dog, it goes around the room looking for a ball. If I remove the ball but leave other toys, which normally the dog seems to find as much interesting as the ball, it still goes looking for the ball and keeps on doing so without showing interest in the other toys.
So the dog isn't prone to pick just anything up in order to get a reward, even in absence of the ball; it must thus know what it's looking for, i.e. the word activated a mental representation of the ball in it's mind.
The notion of language in animals must threaten your sense of uniqueness, but I guess we're not that special.

By RijkswaanVijanD (not verified) on 12 May 2010 #permalink

@7

You haven't addressed my argument at all, and are just reiterating the same claims Mr. Goldman made.

The distinction you're not making is that "indicating certain things" is not enough. In high school, the ringing of the school bell indicated that it was time to go to the next class. But we would not use a ringing sound to talk about how many classes we were taking this year, or how hard this one class was, or how long classes are, as would be the case if the ring actually stood for that concept. That is what actual words (symbols) allow us to do, be they acoustic, visual, haptic, or otherwise. If you try you can think of countless stimuli in your life that you associate with something else (i.e. they activate mental representations of something). I associate The Warrior by Patty Smyth with the feeling of being at my old job and wanting to go home, because I heard that song frequently while working there (and I hated that job). I associate a certain smell with my girlfriend. I associate the sound of my garage door opening with someone coming home. These are all very specific associations, but they are not related to language. Of course, they could be if we so desired. The members of my high school could have started using a ringing sound as the word for "class," and used that in conversations as honest-to-goodness language. But we did not have to have that understanding of "ring" as language in order to understand it as a signal, or indicator of something.

That is the distinction you are not making. If you do not make that distinction, then you have to concede that every time an animal perceives a stimulus and then responds in some accurate, specific way, that animal is demonstrating linguistic abilities. You have to concede that my dog believes that the sound of his leash clanking on the counter is the word for "walk," as opposed to just an indicator that we're going for one (and you have to concede this even though we humans do not consider that sound to be language). And you have to concede that, to a majority of American citizens, the sound of a school bell ringing is a word for "class" or "going to the next class," even though none of us think it is.

At this point I could wrap up my reponse with a snide remark about how you must be emotionally invested in your beliefs and that is why you continue to hold on to them even though they're demonstrably incorrect, but I won't, because I don't care why you believe what you do. Just back it up with evidence that actually addresses the arguments that are being made, instead of reiterating the OP's argument.

@Jason Goldman
Nice talkin to ya.

By Tim Martin (not verified) on 12 May 2010 #permalink

@8: We certainly make associations between various stimuli. Smelling peppertrees makes me think of my summer camp; for a friend of mine, smelling currant makes her think of her summer camp. Does that mean that either of those smells symbolically stands for "summer camp"? I don't think so.

I think the question you are asking is whether or not dogs have symbolic thinking. If they look at a picture of their owner, for example, do they understand that it is not the same thing as their owner, and instead simply represents the person that is their owner. Likewise with the acoustic stream /ball/ - does the dog understand that /ball/ and the actual ball are not truly the same thing? That one is a referent for the other? I am not sure what the answer to this question is (but I will find out, and I will blog it.)

So, yes, the argument that dogs have linguistic comprehension is speculative at best, but its not a speculation that I made (or that the authors of the paper made).

The statement I made (and the authors made) was

this research suggests that the ability to attach meaning to specific sounds may have emerged earlier than, and evolved independently from, the ability to flexibly produce those sounds

and

What appears to be a complex linguistic skill used by human children to learn new words may depend on simpler cognitive building blocks also present in at least one other species, the domesticated dog.

And the evidence does suggest pretty convincingly that at least this specific dog binds different acoustic streams to objects, that it can determine the acoustic stream that should bind to a new object on the basis of exclusion learning, and that it does a relatively good job of remembering those bindings after only one exposure, 4 weeks later.

The bindings of specific sounds to objects - deriving meaning from sounds - does not depend on language. For example, monkeys have specialized calls for different types of predators (see here). The monkeys absolutely attach certain meanings to the different acoustic streams, but nobody would argue that this constitutes language. What it constitutes is deriving meaning from sound, and associating that meaning with an object (or, in this case, another animal).

What I was objecting to in my comments was the idea that studies like this one provide evidence that dogs use stimuli as symbols to represent other things (in other words, that dogs understand stimuli similar to how we understand words). If you didn't mean to make that assertion, than we really have no disagreement. Still, your wording makes it seem like you did. No one talks about monkeys "learning new words" or learning "the principle that objects have labels" when they learn to press a button in response to a red computer screen. So why do we use that language when dogs do something similar, i.e. respond correctly and specifically to an acoustic stimulus?

Anyway, thanks for your response. I realize now that fast mapping and Rico's learning abilities were more the focus of your post, but the symbolic thinking issue also came up and it's one I particularly care about, so cheers.

By Tim Martin (not verified) on 13 May 2010 #permalink