Depths and limits of spontaneous categorization in a family dog.

Categorization has been tested in non-human animals after extensive training procedures under laboratory conditions and it is assumed that in non-primate species categorization relies on perceptual similarity. We report evidence of the ability to categorize objects in absence of specific training in a family dog with vocabulary knowledge of multiple toys, including exemplars of 4 categories. Our experimental design was devised to test categorization in absence of specific training and based on the spontaneously learned vocal labels of the categories, a condition that mirrors human studies more than previous experiments on non-human animals. We also observed that the dog's categorization skills were more accurate when, prior to the categorization test, she was given the opportunity to play with the novel exemplars, suggesting that category representations arise not only from physical resemblance, but also from objects' affordances (function).

Introduction

Categorization is the mental ability to group discriminable objects or events into classes by means of some principle or rule and to respond to them in terms of their class membership instead of as individual stimuli (e.g.[1,2]).

Categorization has been in the focus of research for both human (e.g.[3]) and non-human animals (e.g.[4,5]). In humans, according to most researchers, categories underpin language learning in that they allow labels to be attached to categories of items[6]. For instance, it has been shown that naming different objects with the same name highlights commonalities among them and promotes categorization, even in infants as young as 6 months[7].

Some linguistic competence makes it simpler to test categorization skills (i.e. it becomes possible to “request” the subject to select an object from a verbally identified category, by using its name). However, studies on pre-verbal infants showed that language is not needed for categorization to emerge[2]. Thus, categorization skills are not necessarily restricted to humans, although it seems that language trained apes perform better than non-language trained subjects in complex categorization tasks[8].

Since the seminal work of Herrnstein and Loveland[9] showing that pigeons can discriminate pictures with people from pictures without people, categorization abilities have been studied in several additional non-human animal species. For example, birds (e.g.[10]) and also fishes[11] have been found to categorize stimuli based on perceptual cues, while research regarding mammals has mainly focused on primates (for review see[4,12]).

A typical categorization experiment on non-human subjects involves two phases[1]: training with a set of exemplars from a given category until a predetermined performance is achieved, followed by[2] a test with a mixture of novel exemplars from the same category and non-exemplars. The degree to which subjects correctly sort these test items into category members versus non-members is considered evidence for categorization performance. While the ability to categorize is assumed to be of obvious advantage in the wild, as it reduces the efforts needed to acquire an appropriate response to different stimuli belonging to the same category, most previous studies on categorization abilities were performed under artificial conditions (i.e., in the laboratory), after massive training of the subjects to reach a predetermined criterion (e.g., Bovet and Vauclair[13] trained baboons on 1460 trials to criterion; D’Amato et al.[14] trained cebus monkeys on more than 1000 trials to criterion; Range et al.[15] trained dogs on 720 to 2040 trials to criterion before starting the categorization of items). Studies about spontaneous categorization skills, in absence of training and under natural (i.e., non-laboratory) conditions are lacking.

It is commonly assumed that, while humans take into account the function of objects, besides perceptual cues, to form mental categories (e.g.[16,17]), non-human animals tend to categorize based on the perceptual features of the stimuli[4]. However, non-human animals are tested after intensive training procedures, during which the rules to be followed for categorization are established by the experimenters, typically based on perceptual features. This may strongly determine the principles used by the animals to categorize novel exemplars during the tests. Moreover, training and testing in the case of non-human animals often involves the use of photographs of the exemplars[11], therefore the possibility to learn about the objects’ affordances is lacking or very limited.

Little is known about the role of objects’ function in non-human animals’ categorization. Some categorization based on affordances has been shown to extend not only to humans, but also to non-human primates (4 for review). Baboons, for example, after massive training, were shown to categorize “food” and “non-food” items[13], thereby challenging the assumption that this type of categorization is linked to language[18]. Furthermore, studies on tool-related cognition in New Caledonian crows (e.g.[19]) suggest that the ability to categorize based on object affordances might be more phylogenetically widespread than commonly thought. However, it is not known whether object affordances may guide categorization in non-primate species or whether their categorization skills are constrained only by perceptual similarity.

Despite dogs having gained importance as subjects of cognitive studies, relatively little research has been conducted on their categorization skills. After being trained to distinguish dog vocalizations from non-dog sounds, dogs were found to discriminate novel (i.e., not trained) sound stimuli belonging to dogs from sound stimuli not belonging to dogs[20] and, after being trained on pictures of dogs and pictures of landscapes, dogs were observed to discriminate novel pictures belonging to these two categories[15]. Dogs were also recently shown to identify novel odors as members of a trained odor category[21]. These studies showed that dogs, after being specifically trained on a given set of stimuli, are able to categorize novel samples based on perceptual similarity (visual, auditory or odor). Chaser, a Border collie that had been extensively trained to learn over 1000 object-names, was also successfully trained to learn common nouns that refer to categories, including discriminating objects with which he was allowed to play from objects that he was not allowed to interact with[22]. Importantly, it should be considered that Chaser’s categorization skills were tested after the dog had been specifically trained to learn and generalize object categories until a predetermined performance was achieved. Thus it is not known whether and how dogs would spontaneously categorize, in absence of specific training.

Among the perceptual cues that humans use to categorize, color seems to play a major role and it has been suggested that categorization based on color might be an innate predisposition[23]. Color categorization has been found in several animal species as well (e.g., zebra finches[10]; fishes – e.g., triggerfish[11]; Chimpanzees[24]). Dogs have only two types of cone photoreceptors in their retina (those allowing perception of blue and yellow) therefore, they have a dichromatic color perception[25,26]. However, it has been claimed that dogs can also distinguish between red and green, at least under some brightness conditions[27]. It has been shown that dogs rely preferentially on color cues (blue and yellow), rather than brightness, in a two-way choice task[28]. However, little is known on whether dogs would spontaneously rely on color to categorize objects. We suggest that color is unlikely to be a fundamental, biologically relevant feature of objects for dogs (see also[29]) and, consequently, we would not expect them to spontaneously use color to categorize when other more relevant features can be used.

To investigate spontaneous categorization skills, we tested a family dog named Whisky, a 4-year old female Border collie with extensive vocabulary knowledge of multiple toy names spontaneously acquired during unplanned playful interactions with her owners. Among her named toys, Whisky also had some exemplars of toys belonging to one of 4 categories (balls, ropes, rings and Frisbees) and toys of different colors (blue, yellow, red, green, orange). The aim of this study was to test whether, despite the lack of specific training, the dog would be able to categorize novel exemplars of objects belonging to those categories, since previous studies on categorization typically relied on extensive training of the subjects in laboratory conditions. This allowed us to test whether non-human animals may spontaneously form mental categories, when exposed to their exemplars in a natural context.

Additionally, we aimed to explore whether increased attention to the objects and experience with their affordances, beside perceptual features (shape), play a role in the dog’s spontaneous categorization tendency. In order to achieve this, the subject was exposed to the novel toys in two conditions, before testing her ability to categorize novel exemplars belonging to those categories: (1) Exposing her to a brief play session with the toys before the categorization test (Play condition) and (2) Without this previous play interaction, but only allowing the dog to explore the toys before the test (Exploration only condition). In every categorization test of every condition, a novel set of toys was used.

We hypothesized that the natural, unplanned exposure to the objects belonging to the different categories and to their verbal labels received during her life in a human family allowed the dog to spontaneously form some mental categories of those objects. We also wanted to gain some insight on whether a brief experience of the typical type of play with the toys belonging to different categories – i.e., the way one plays with balls or ropes etc. – and the consequent increased attention to the objects, would enhance the dog’s accuracy in categorizing the novel exemplars. Additionally, we tested whether the dog had formed some color category by testing her ability to select familiar toys based on their color.

Results

Baseline: retrieving known toys based on their verbal labels (“names”)

We first tested Whisky’s knowledge of the proper name of all the toys available at her house, to which her owners had given a name (59 available toys, including also the toys of the 4 categories) by asking Whisky to fetch all of them, one by one, upon hearing their names.

Before the test started, 20 randomly chosen toys were laid on the floor of the living room by the experimenter – out of view from the owner who stayed in the kitchen for the whole duration of the tests. During the test, the owner requested the toys in a predetermined randomized order by saying: “Bring