Dogs don't generalize words like humans do

When humans learn the names of objects, they generally associate that name with the overall shape of the object. Thus, if they are taught that a U-shaped object is called a ‘DAX’ and asked to identify another DAX, they’ll pick an object with the same shape, but not necessarily the same size or texture. In other words, when it comes to nomenclature, we generalize in a specific way with a bias toward shape. Is this true for dogs as well?

There are a handful of dogs with a large enough vocabulary to test this feature of language. I’ve written before about Chaser, who knows as many words as a human toddler (over a thousand). This time, the test subject was a Border Collie named Gable, who reliably knows the names of over 40 different objects.

Emile van der Zee, Helen Zulch and Daniel Mills from the University of Lincoln and Gable's trainer taught the dog specific nonsense names for some novel objects. For example, in one set of experiments, he was taught that the L-shaped object shown as (1) in the figure below was a GNARK. When asked to retrieve another GNARK from amongst pairs of the objects shown, he invariably chose something the same size (2 or 3) rather than the same shape (4 or 7). Apparently, Gable was not making the same mental associations that humans do.

doi:10.1371/journal.pone.0049382.g003

I don’t know about you, but I’d say that if object 1 was presented to me as a GNARK, then objects 4 and 7 are definitely GNARKS as well. This may be because, as a human, I rely heavily on my eyesight to interpret the world. Dogs no doubt rely on other cues, perhaps weight or mouth-feel. As you can see below, Gable does mouth each object before deciding which is the GNARK. Since word generalization is a critical feature in language, these data may mean that the way humans developed language is not the only road to that achievement. 

Below, a test session in which Gable is asked to retrieve a u-shaped DAX. Between trials, the trainer lets Gable play with the DAX prototype. As the test begins, the researcher pretends to make that prototype one of the choices.

Test session in which Gable is asked to retrieve a u-shaped DAX. Between trials, the trainer lets Gable play with the DAX prototype. As the test begins, the researcher pretends to make that prototype one of the choices.



Emile van der Zee, Helen Zulch, & Daniel Mills (2012). Word Generalization by a Dog (Canis familiaris): Is Shape Important? PloS ONE : doi:10.1371/journal.pone.0049382