Fetching what the owner prefers? Dogs recognize disgust and happiness in human behaviour
Research using the two-object choice paradigm showed that dogs prefer the object associated with the happy human emotion. However, they provided rather ambiguous results regarding the negative emotions. We assumed that differences between the dogs’ and owners’ interest towards the ‘negative’ object might be responsible for this. In our experiment, dogs observed their owner expressing different emotions towards two uniform plastic bottles. Five dog groups were tested based on the condition they received: (1) happy versus neutral, (2) happy versus disgust, (3) neutral versus disgust and (4–5) neutral vs neutral, as control groups. Contrary to previous studies using free choice paradigm, we used a task-driven approach. After the demonstration, the dogs had to retrieve one object to the owner. The dogs’ performance in the two neutral–neutral groups did not differ from the chance level. In contrast, subjects were able to distinguish between the happy and neutral expression of the owner: they both approached and fetched the ‘happy’ object. In the happy–disgusted and neutral–disgusted groups, the dogs approached the bottles randomly, suggesting that they found the ‘disgusting’ and ‘neutral’ objects equally attractive. Nevertheless, the dogs preferentially retrieved the object marked with the relatively more positive emotion (happy or neutral) to the owner in both conditions. Our results demonstrate that dogs are able to recognize which is the more positive among two emotions, and in a fetching task situation, they override their own interest in the ‘disgusting’ object and retrieve what the owner prefers.
In the last 2 years, several studies investigated dogs’ ability to discriminate between human facial expressions or between different tones of voices. Deputte and Doll (2011) showed still facial expressions of the experimenter to dogs, and they found that subjects reacted more to the facial expressions of anger and joy than to neutral faces. Nagasawa et al. (2011) reported that dogs can discriminate between photographs of smiling and blank faces of their owners. In contrast, Hori et al. (2011) found no difference in the dogs’ looking time at the photographs of their owners’ smiling, angry and neutral expressions. Regarding the acoustic modality, the results are more contradictory. Dogs can discriminate between emotionally different tones of voice (Ruffmann and Morris-Trainor 2011), that is, they were slower to take a piece of food when commanded to leave it in an angry tone of voice compared with a ‘happy voice’. In contrast, Mills et al. (2005) found no difference in the latencies to obey when the ‘sit’ and ‘come’ commands were given in different emotional tones.
It seems that dogs do acquire some information from the human face and voice about our emotional states. However, most of the communicative interactions between owners and dogs involve simultaneous visual and vocal signals, thus investigating only one modality may not be representative of the dogs’ general ability to interpret human emotional expressions.
Buttelmann and Tomasello (2013) introduced an experimental paradigm (based on Repacholi 1998) in order to test whether dogs are able to rely on the emotional behaviour of humans in a two-object choice task. They allowed dogs to select one of two boxes after viewing the experimenter’s emotional reaction to these boxes (looking into the boxes with different facial expressions accompanied by verbalizations: happy versus neutral, happy versus disgust). Each dog participated in 18 trials of both conditions. The dogs chose the ‘happy’ box above chance level in the Happy–Disgusted condition, but they failed in the Happy–Neutral condition (similar to great apes, Buttelmann et al. 2009). Buttelmann and Tomasello (2013) claimed that the dogs’ failure in the Happy–Neutral condition could be either due to difficulties in distinguishing between the happy and neutral emotions, or the dogs’ assumed negative affective response to the neutral expression.
Merola et al. (2014) addressed the hypotheses that neutral expressions may have negative effects by including a novel negative-neutral condition. They used a similar experimental setup as Buttelmann and Tomasello (2013). Dogs could choose between two identical objects marked with different emotional expressions (happy versus neutral, happy versus fearful, neutral versus fearful). Each dog participated in one trial only. Dogs were able to distinguish between the happy and fearful expressions, but only if the owner was the demonstrator (the dogs’ choice was random when a stranger demonstrated the same emotions). They also found that dogs distinguished between the happy and neutral expressions of their owner (preferring the happy one, contrary to the results of Buttelmann and Tomasello 2013). However, dogs chose randomly between objects marked with fearful versus neutral expressions. The authors concluded that in such situations, dogs have a tendency to show a ‘preference for the positive emotion’ (rather than ‘avoidance of the negative emotion’). The lack of preference in the neutral–fear condition could be either due to the dogs’ inability to recognize the valence of the fearful expression, or due to lack of inhibition of exploratory behaviour in response to human fear.
The latter explanation may reflect a possible difference between the preferences of the owner and dog. In the case of positive emotions, the interest and preference of the owner and dog usually match (i.e. the owners often use happy, excited emotions when trying to get the dogs’ attention, e.g. in playing or training situations, so what the owners show preference for is usually also interesting for the dog). However, in the case of negative emotions, the interest of the dog and the owner could be opposite. In everyday life situations (e.g. during walks), what the owner finds negative (e.g. disgusting) could be interesting for some dogs (e.g. garbage, faeces). If the dogs are able to recognize the valence of the owners’ negative emotional expressions, some dogs may have learnt to associate it with a negative outcome and avoid such objects, while for other dogs, the owners’ negative emotions may mean a rather interesting object, which elicits an approaching behaviour.
In other words, in a free choice situation (Buttelmann and Tomasello 2013; Merola et al. 2014), dogs may have recognized the valence of the demonstrator’s negative emotional display and understood the link between this emotion and the object, but some dogs were willing to ignore this information because it was inconsistent with their own preference, resulting in a random choice at the group level. Based on this reasoning, we suppose that analysing only the dogs’ approaching behaviour towards an object does not reliably reflect their ability to recognize the valence of human emotions, as it is influenced by their interest towards these objects. To analyse whether the dogs are able to recognize the negative emotional signals of the owner, as well, the dogs should interpret the demonstration as a task situation. By giving the dog a command used in play situations (‘Fetch!’), the situation of choosing an object became an interactive play task instead of a non-interactive free choice. As the dog can only play fetch-and-carry with a partner, the partner’s (here, the owner’s) preference became a more relevant factor influencing the dog’s choice of object (for example, in an everyday play situation: which toy the owner wants to play with, or which stick the owner had thrown). Accordingly, we hypothesized that the owner’s demonstration is more relevant in an interactive task situation than in a non-interactive free choice task. Therefore, if we ask dogs not only to approach a chosen object but to retrieve one to the owner, we can distinguish between the dogs’ own preference and the ability to recognize human emotions.
In our experimental setup, dogs had to choose between two similar objects (plastic bottles) which were associated either with a positive, neutral or negative emotional expression. The bottle associated with the more positive emotion contained food, and the other bottle contained a small stone. We decided to use the owner as the demonstrator because dogs are more familiar with his/her emotional expressions and more prone to rely on it (Merola et al. 2014). As a negative emotion, we used disgust (similar to Buttelmann and Tomesello 2013), because it may be more frequently expressed (even over-emphasized) by the owners in the dogs’ everyday life so that the dogs may have more opportunity to learn the association between this expression of the owner and a negative outcome (e.g. scolding) than in the case of fear.
In the present study, each dog participated in eight experimental trials to investigate consistency in choice behaviour. Following the methods used by Merola et al. (2014), each dog in our study received only one pair of emotional displays (happy versus neutral, happy versus disgust, neutral versus disgust or neutral versus neutral), because a pilot study showed that the performance in the first condition strongly affected the dogs’ choice in the subsequent conditions. To maintain the dogs’ motivation to choose one of the objects, after a bottle was fetched, we opened it and showed its contents (food or stone) to the dog in every trial. If it was the food pellet, the dog was allowed to eat it. We included two control conditions, one to investigate the possible confounding effects of odour cues and another to investigate the possible ‘Clever Hans’ effect. In these conditions, both the baited and non-baited bottles were associated with neutral facial expressions.
The key important difference in comparison with earlier studies was that dogs were instructed to retrieve an object to the owner, not only to approach it. This protocol allowed the dogs to approach any of the objects presented but then to choose freely which one they preferred to retrieve. In this way, we could obtain a measure of the dogs’ own preference (first approach) and their tendency to recognize the valence of the human happiness and disgust emotions (specific bottle fetched).
We hypothesized that dogs will preferentially choose (approach and fetch) the object marked with the happy emotion over the other one marked with a neutral expression since here the interests/preferences of the owner and dog match. In case of the disgust emotion, we hypothesize that some dogs may display interest towards the object that the owner finds disgusting, while other dogs do not. Thus, we expect a random first approach at the group level in the Happy–Disgusted and Neutral–Disgusted groups. However, as some dogs may also have learnt to associate retrieving ‘disgusting’ objects to the owners with a negative consequence, we expect that the dogs will avoid retrieving the ‘disgusting’ object to the owner.
Merola et al. (2014) assumed that the previous experience and learning influenced the dogs’ choice behaviour in such object choice tests. That is, during their ontogeny, dogs had learnt the association between the owners’ happy, enthusiastic display and a positive outcome, and therefore, they show preference for the object marked with the positive (happy) emotional display. On the other hand, the dogs’ skills for reading human social communicative behaviour (i.e. recognize certain human emotional displays) might also be the result of the genetic changes caused by the domestication (e.g. Hernádi et al. 2012; Miklósi et al. 2004). However, no study yet investigated the performance of non-adult puppies in emotion-recognition tasks. Here we tested a small number of puppies, as well, to compare their performance with that of the adult dogs.
In sum, the aim of this study was to investigate (1) whether dogs are able to discriminate the human happiness and disgust emotional expressions from each other and from the neutral one and (2) whether they prefer the object eliciting the more positive emotion from the owner in a two-object choice test. Compared with previous studies, the unique aspects of our experiment were that we also took into account the dogs’ interest towards the negative, ‘disgusting’ object by analysing both the object first approached and the one they retrieved to the owner and that we also investigated the performance of puppies.
Materials and methods
A total of 125 adult (>1 year) pet dogs and 38 puppies (2.5–10 months old) from various breeds were recruited on a voluntary basis from the Family Dog Project database in Budapest, Hungary. The only criterion for inclusion was that dogs had to be familiar with the ‘Fetch’ command. Fourteen adults and ten puppies were excluded for various reasons (e.g. the dogs lost their interest, or owners failed to follow our instructions), and an additional 12 dogs (11 adults, one puppy) were excluded from the analyses due to 100 % side preference (when the dog chose the object placed on one side in all the trials).
The remaining 127 dogs (adults: 42 males, 58 females; mean age ± SD = 3.74 ± 2.34 years, puppies: 17 males, 10 females; mean age ± SD = 0.50 ± 0.19 years) belonged to 39 different breeds, and 26 dogs were mixed-breed. Data of all the dogs included in the study are provided in Online Resource 1. The 100 adult dogs were semi-randomly assigned to five groups (three experimental and two control groups, 20 dogs in each) based on the emotion pair they received. As only a small number of puppies were available (N = 27), they were distributed only among the three experimental groups (9 puppies in each). We assumed no difference between puppies and adult dogs in their ability to sniff out the food in the control conditions, and also no difference were expected in their owners’ motivation to provide ‘Clever Hans’ cues.
Objects and testing room
A video of the protocol can be seen in Online Resource 2 and on the Comparative Mind Database: http://www.cmdbase.org/web/guest/play/-/videoplayer/223.
Dogs were free to explore the room prior to the testing for 5–6 min. The test started with warm-up trials. The aim of these trials was to practice retrieving a bottle to the owner on only a verbal command. The owner sat in a chair and held the dog. When the subject was watching, the experimenter put a piece of food in a plastic bottle (similar to those used in the test trials), closed the bottle then put it down 1 m from the dog. The owner then encouraged the dog to retrieve the bottle; then the owner gave its content to the dog. This procedure was repeated until the dog retrieved the bottle upon the first command.
Each test trial was executed in exactly the same way:
Baiting phase: The owner sat down on the chair and put the dog on leash. The experimenter turned her back to the subject, baited the bottles, and put them in their predetermined locations one by one, in a random order. Then she returned to the owner, took the leash of the dog, and instructed the owner about the setup of the following demonstration (starting side and the order of the emotions) (Fig. 2a).
Demonstration phase: The owner stood up, attracted the dog’s attention if necessary, and walked to the first bottle. Then she/he turned back to the dog, crouched down behind the bottle, touched it, looked at the dog, and gave the appropriate emotional expression (happy, neutral or disgust) for 3–4 s (Fig. 2b). Then the owner put the bottle back in its place, walked to the other object, and repeated this display with the second assigned emotion. During the demonstration, the experimenter stood silently behind the dog, looking towards the middle of the bottles. After the demonstration, the owner walked back to the chair, sat down, and positioned the dog in the middle.
Fetching phase: If the dog assumed the predetermined body position, then the owner released it, and immediately gave the ‘Fetch!’ verbal command. The owner was strictly instructed not to use any gestures or directional cues, and they were required to look straight ahead between the bottles while giving the command. If the dog started to move towards the bottles, the owner stopped talking and sat silently and motionless. When the dog retrieved one of the bottles to the owner, it was briefly praised (irrespective of whether the baited or the non-baited bottle was retrieved), and then the owner got the food/stone out of the bottle, and offered it to the dog (allowed it to eat the food or smell the stone). During this phase, the experimenter stood silently next to the owner, looking at a point halfway between the bottles. Next, the experimenter retrieved both bottles, and the next trial started with the hiding phase.
Each dog received eight trials, the side of the bottle containing food changed in every trial, and the direction of the demonstration (from left to right or vice versa) changed in every second trial. The owners’ starting side in the first trial was counterbalanced among dogs. Each dog was pseudo-randomly assigned to one of the five experimental groups:
Happy–Neutral group (N = 20 adults, 9 puppies): the owner reacted to one of the bottles with a happy emotional display (this bottle contained the food) and with a neutral display to the other bottle (this one contained the stone).
Happy–Disgusted group (N = 20 adults, 9 puppies): the owner reacted to one of the bottles with a happy display (contained food) and with a disgusted expression to the other bottle (contained stone).
Neutral–Disgusted group (N = 20 adults, 9 puppies): the owner reacted to one of the bottles with a neutral display (contained food) and with a disgusted expression to the other bottle (contained stone).
Neutral–Neutral (control) group (N = 20 adults): the owner reacted with a neutral expression to both bottles; one of them contained food (the owner was not aware which) and the other a stone. This condition served as an odour control group, included in order to investigate if the dogs are able to smell the location of food and choose it irrespective of the owners’ demonstration.
Clever Hans control group (N = 20 adults): similar to the group above, the owner reacted with a neutral expression to both bottles; one of them contained food and the other a stone. In this group, the owners were told that the aim is to test whether the dogs are able to sniff out where the food is. The owners were informed about the location of the baited bottle after each demonstration right before they let the dog go. The experimenter also added comments, which may have elicited some kind of expectation in the owner, like ‘I hope the dog will find the food this time’. This condition served as control group, included in order to investigate if the dogs’ choice are influenced by the owners’ voluntary or involuntary ‘Clever Hans’ cues during the fetching phase.
The owners expressed happiness or disgust emotions by displaying facial and body gestures accompanied by verbalizations. The reason behind using the owner as the demonstrator was that dogs are supposedly more familiar with their owners’ emotional expressions (Merola et al. 2012, 2014). The owners were instructed that they should try to behave as they usually do while displaying these emotions. For example, they were instructed to act as if they were trying to invite the dog to play in case of the happy emotion and imagine that their dog found something particularly distasteful during walking in the case of disgust. They were also encouraged to use vocalization, but they were not allowed to use any word known as a command for the dog during the demonstration. The neutral emotion was displayed by only a blank facial expression; here, no vocalization was allowed.
In the Demonstration phase, we evaluated the owners’ behaviour at the bottles. We coded the length of the emotional display, the percentage of talking to the dog, looking at the dog and touching the bottle, and the frequency of pushing the bottle away and pulling the bottle closer in one randomly chosen trial for each dog. We compared these variables between the three emotions (happy, disgust and neutral) using one-way ANOVA. Moreover, we also investigated whether the owners demonstrate the same emotion differently in different conditions. For this, we compared the conditions in which a given emotion was demonstrated (e.g. the demonstration of the ‘happy’ emotion in the Happy–Neutral and Happy–Disgusted conditions) using independent-sample t tests. We also investigated whether the owners display the neutral emotion differently at the bottle containing food than at the bottle containing stone in the two control conditions using paired-sample t tests.
In the Fetching phase, the trials were scored on the spot by the experimenter (B.T. or F.Sz.), but all experiments were recorded on video, as well. We measured two variables in each trial: the first approach (corresponds to the object the dog first touched in a given trial) and the fetched bottle (the object the dog retrieved to the owner). Both variables were categorized as correct (the object contains the food) or incorrect (the object contains the stone). A randomly selected 25 % of the subjects were recoded to assess the inter-observer agreement between the two experimenters. The agreement was perfect between them (Cohen’s Kappa = 1.00 for both variables).
IBM SPSS Statistics v21 was used for statistical analyses. We analysed whether the dogs’ performance was affected by the condition they received, the order of the emotional expression (demonstrated first or second), the spatial location of the object (left side or right side), the repetition of the trials (first four vs. second four) or the age category (adult or puppy). For these, we used two binary generalized linear mixed models (GLMM), one for the first approach, and one for the fetched bottle variables. In each model, the dogs’ choice in each trial (correct or incorrect) was added as the target variable, and the condition, the demonstration order, the side of the baited bottle, the repetition (belongs to the first half or to the second half of the trials) and the age category were added as fixed effects. Two-way interactions between the condition and order, condition and side, and condition and repetition were also investigated. Non-significant effects were removed from the models. If the condition the dogs received was found as a significant predictor of their performance, we compared the performance in each group to chance level (50 %) using one-sample Wilcoxon Signed Rank tests.
We also analysed the effect of learning during the trials by analysing the dogs’ performance in the first trial (which is free of any possible learning effect) using one-tailed Binomial tests.
Descriptive statistics of the emotional displays of the owner during demonstration
Variable (mean ±SD)
Talk time (%)
Look at dog (time %)
Touch bottle (time %)
Push away bottle (N)
Pull close bottle (N)
Length of display (sec)
Happy (in N = 58 trials)
43.7 ± 24.9
85.6 ± 8.7
0.4 ± 0.5
7.8 ± 2.9
75.2 ± 16.3
48.6 ± 26.3
88.0 ± 9.0
0.4 ± 0.8
7.2 ± 2.5
Disgust (in N = 58 trials)
62.0 ± 22.1
37.0 ± 31.3
72.2 ± 25.8
0.5 ± 0.6
0.1 ± 0.4
6.7 ± 2.5
53.5 ± 23.7
19.4 ± 19.1
65.7 ± 26.1
0.4 ± 0.6
0.1 ± 0.3
6.4 ± 2.5
Neutral (in N= 98 trials, 138 displays)
26.3 ± 24.2
83.2 ± 13.4
5.7 ± 2.0
28.1 ± 25.1
82.2 ± 18.7
0.04 ± 0.2
6.6 ± 3.0
26.8 ± 21.2
76.8 ± 9.7
0.1 ± 0.5
5.5 ± 1.8
26.8 ± 18.9
76.1 ± 13.4
0.1 ± 0.2
5.7 ± 1.7
Clever Hans (food)
20.8 ± 22.7
81.3 ± 7.2
0.1 ± 0.3
5.3 ± 2.7
Clever Hans (stone)
14.0 ± 17.7
81.5 ± 6.6
0.1 ± 0.3
4.8 ± 2.1
We also compared the demonstration of a given emotion between the conditions it emerged. No difference was found in the owners’ demonstration of the happy emotion between the Happy–Neutral and Happy–Disgusted conditions. In the case of disgust, the owners looked more at the dog in the Happy–Disgusted condition than in the Neutral–Disgusted condition (P = 0.018); no other difference was found between the two conditions. In the case of the neutral demonstration, no difference was found between the Happy–Neutral and Neutral–Disgusted conditions. Similarly, no difference was found in either of the control conditions (Neutral–Neutral and Clever Hans) between the demonstration at the baited bottle and the demonstration at the bottle containing stone.
Fetching phase: the effect of the condition, the order of the emotional expressions and the location of the object
Parameter estimates of each fixed effects in a) first approach and b) fetched bottle variables
a) First approach
Condition (reference category: Neutral–Neutral (control) group)
Clever Hans control group
Demonstration order (reference category: firstly demonstrated)
Secondly demonstrated bottle
Side of the baited bottle (reference category: left side)
Right side bottle
Repetition (reference category: first half of the trials)
Second half of the trials
Age group (reference category: adults)
b) Fetched bottle
Condition (reference category: Neutral-Neutral (control) group)
Clever Hans control group
Demonstration order (reference category: firstly demonstrated)
Secondly demonstrated bottle
Side of the baited bottle (reference category: left side)
Right side bottle
Repetition (reference category: first half of the trials)
Second half of the trials
Age group (reference category: adults)
Neither the demonstration order, nor the side on which the object was placed, nor the repetition (first half vs. second half of the trials), nor the age category had a significant effect on the dogs’ performance (first approach and fetched bottle; P > 0.355 for all). No significant interaction with the condition was found, either. The condition itself had a significant main effect on both the dogs’ first approach (F4,982 = 2.433, P = 0.046) and on which bottle the dogs retrieved to the owner (F4,982 = 3.482, P = 0.008). Pairwise contrasts revealed differences between the control groups (Neutral–Neutral and Clever Hans) versus the three experimental groups in both variables.
Neutral–Neutral (control) group: In the first approach, the performance in this group differed significantly from the Happy–Neutral group (P = 0.035) and a nearly significantly from the Neutral–Disgusted group (P = 0.083). In the fetched bottle variable, all three groups had significantly higher performance than this group (Happy–Neutral P = 0.002; Happy–DisgustedP = 0.036; Neutral–DisgustedP = 0.046).
Clever Hans control group: In the first approach, the performance in this group differed significantly from the Happy–Neutral and Neutral–Disgusted groups (P = 0.009; P = 0.027, respectively) and tended to differ from the Happy–Disgusted group (P = 0.095). In the fetched bottle variable, this group had significantly lower performance than the Happy–Neutral and Happy–Disgusted groups (P = 0.004; P = 0.049, respectively), the Neutral–Disgusted group had only a marginally higher performance than this group (P = 0.060).
No differences between the Neutral–Neutral (control) group and Clever Hans control groups were found (first approach: P = 0.654; fetched bottle: P = 0.911).
These results showed that the condition seems to be a significant predictor of the dogs’ choice behaviour, so the performance in each group was also assessed separately. Since there were no significant differences in performance of puppies and adult dogs, their data were combined for these analyses.
Fetching phase: performance in each group
In the Clever Hans control group, the dogs’ approached the baited bottle in 46.3 % of the trials and retrieved this bottle to the owner in 51.9 % of the trials. Again, no difference from the chance level was found (first approach: T + = 24.00, N = 12 (8 ties), P = 0.266; fetched bottle: T + = 46.00, N = 12 (8 ties), P = 0.622).
In the Happy–Neutral group, the dogs’ performance differed from chance level regarding both variables. They approached the ‘happy’ object first in 59.8 % of the trials (T + = 228.50, N = 23 (6 ties), P = 0.004) and retrieved this bottle to the owner in 66.6 % of the trials (T + = 293.00, N = 24 (5 ties), P < 0.001).
In the Happy–Disgusted group, the dogs’ first approach did not differ significantly from chance level, only a nearly significant effect was found. In 55.6 % of the trials, the dogs approached the ‘happy’ bottle (T + = 130.00, N = 18 (11 ties), P = 0.054). However, they retrieved the ‘happy’ bottle significantly above chance level (62.2 % of the trials, T + = 242.50, N = 24 (5 ties), P = 0.007).
The same pattern emerged in the Neutral–Disgusted group, the dogs’ first approach only nearly significant: in 57.5 % of the trials, the dogs approached the ‘neutral’ bottle (T + = 231.50, N = 25 (4 ties), P = 0.063), but they retrieved the ‘neutral’ bottle to the owner significantly above chance level (62.3 % of the trials, T + = 252.00, N = 23 (6 ties), P < 0.001) (Fig. 3).
Fetching phase: performance in the first trial (effect of learning during the trials)
Our study aimed to investigate whether dogs recognize and rely on the owners’ emotional expression of happiness and disgust in a two-object choice test, taking into account the dogs’ curiosity and interest towards the objects. We hypothesized that dogs do recognize both the positive and negative valence of owners’ emotions, but that the object first approached is strongly influenced by their interest towards these objects, whereas the object they retrieve to the owner is also influenced by the owners’ preference. Therefore, the object which is first approached by the dog could be different than the one which is retrieved to the owner. We expected that in a task situation, dogs would retrieve the ‘positive’ object and avoid retrieving the ‘negative’ object to the owner.
Our results showed that dogs recognized the valence of the owners’ positive and negative emotional displays and similar to other human communicative cues (e.g. pointing, gazing); they are able to use it as a source of information. However, the mean performance in the three experimental groups (57.6 % in first approach, 63.7 % in fetched bottle) was lower than in other two-object choice tasks (e.g. distal pointing: ~80 % in Lakatos et al. 2009). The reason behind the lower performance in this experiment might be attributed to the 5–10 s delay between the demonstration and the choosing phase. Previous studies (e.g. Fiset et al. 2003; Topál et al. 2005) found that delay before the choice can cause a decline in the dogs’ performance in object choice tasks. Another, not mutually exclusive explanation could be the effect of the local enhancement. In most of the two-object choice tasks (e.g. pointing, gazing.), only one object is marked with cueing, while in our experiment, the owners provided highly salient social cues at both objects (e.g. touched it while looked at the dog). Since dogs are sensitive to such cues (e.g. Téglás et al. 2012), demonstrating them at both objects could slightly mask the difference in the content of the demonstration and also make the two objects more similar in memory.
Dogs in the Happy–Neutral group approached and retrieved the ‘happy’ object to the owner above chance level, and the performance (both the first approach and the fetched bottle) in this condition differed from that in the Neutral–Neutral (control) and Clever Hans control groups. These results support previous findings (Merola et al. 2014) that dogs recognize the valence of the happy emotion, and they preferentially choose the indicated object over the other one marked by a neutral behavioural expression. However, our result contradicts findings reported by Buttelmann and Tomasello (2013) where the dogs chose randomly when the experimenter displays happy and neutral emotions. One reason behind this contradiction could be that the dogs were familiar with the owners’ emotional displays, but not with that of the experimenter (as suggested by Merola et al. 2014). Alternatively, the discrepancy can be attributed to differences in the design of the studies. Both the present study and the study by Merola et al. (2014) exposed each dog to only one pair of emotional displays, whereas Buttelmann and Tomasello (2013) used a within-subject design. The lack of preference in the latter study could be explained by the fact that half of the dogs participated in the Happy–Neutral condition after the Happy–Disgusted condition. These dogs might be more inclined to investigate the ‘neutral’ object, because in this case, they were not firmly discouraged (i.e. with the disgusted emotional expression in the Happy–Disgusted condition) to do so.
In case of the disgust, we hypothesized that some dogs may be predisposed to display interest towards the object that the owner finds disgusting. Thus, we expected random performance in the dogs’ first approach in the Happy–Disgusted and Neutral–Disgusted groups at the group level. However, by putting the dogs in a task situation, we predicted that they would avoid retrieving the ‘disgusting’ object to the owner in both conditions.
The results confirmed these predictions. In the Happy–Disgusted and Neutral–Disgusted groups, the first approach of the dogs did not reached the significant level, suggesting that the dogs at the group level seem to be nearly as interested in investigating the object eliciting disgust from the owner as the object eliciting neutral or happy displays. However, contrary to their first approach, the dogs retrieved the bottle marked with the more positive emotion (happy or neutral in contrast to disgusted) significantly above chance level. The performance in the fetched bottle variable in both groups differed from that in the Neutral–Neutral (control) and Clever Hans control groups. It seems therefore that dogs are able to distinguish between the disgusted and neutral emotional expressions of their owners and are able to recognize the valence of the disgust, as well. Importantly, significant avoidance of the ‘disgusting’ object emerges only in a task-driven situation.
As a simple explanation, during everyday life, family dogs may have learnt to associate fetching objects the owners find disgusting with a negative outcome, and as a consequence, they avoided retrieving the ‘disgusting’ bottle to the owner. However, we found no difference in the performance between the adults and puppies. Thus, one can argue that the ability of recognize human emotional signals could have also evolved during the process of domestication, similar to dogs’ other specific social skills (Miklósi et al. 2004), as it might be a very useful tool for dogs to adapt to the human society.
Our findings are similar to those reported for human infants and great apes. In the study of Repacholi and Gopnik (1997), 14- and 18-month-old infants viewed the experimenter’s emotional reactions (happy versus disgusted) to two types of food, one of which was preferred by the infants. Then they were asked to give the experimenter a piece of food from the two bowls. Infants at the age of 18 months offered the food type the experimenter preferred both when it matched their own preference (76 %) and when it did not (69 %). In the experiment of Buttelmann et al. (2009), apes viewed the experimenter’s emotional reaction (happy versus disgusted) to two containers (containing different types of food) and then watched the experimenter eating something from a container. Then they were allowed to select a container for themselves. If the apes did not know about the contents of the containers (so their choice was not influenced by their own preference), then they showed a slight preference (56 % of the trials) for the cup still containing food (i.e. the cup which elicited disgust from the experimenter). Based on these studies, one might speculate that dogs are also able to differentiate between what they themselves find interesting and what their owners prefer. The dogs first approach what they themselves prefer, but they infer the owner’s desire and then retrieve the object which the owner showed preference for during the demonstration.
The highest performance was found in the Happy–Neutral group in both variables. It suggests that the susceptibility to recognize human emotions might be emotion specific. The dogs may be more predisposed to recognize those human emotional displays, which show more generality across species (like joy–happiness, e.g. Ekman 1992; Morris et al. 2008), while they need more time to learn to recognize emotions, which are more human specific (such as disgust, Rozin et al. 1999) (although, no difference between the adults and puppies performance was found). Second, dogs might be generally more exposed to the owners’ happy displays in everyday life situations, since it may occur more frequently than disgust when interacting with the dog (e.g. during playing, training, or just petting). Therefore, dogs have had more opportunities to learn to associate the human happy expression with a certain outcome (although, again, no difference between the adults and puppies performance was found). Third, as mentioned in the introduction, the interests of the owner and dog are more likely to match in the case of positive emotions, resulting in a higher performance (both in first approach and fetch) in conditions where only the happy emotion is involved. Fourth, one may argue that differences in the salience of the demonstrations can also emerge as an alternative explanation for the higher performance in the Happy–Neutral group (i.e. more salient cues in the happy emotional display than in the neutral one, see Table 1). However, in the Neutral–Disgusted group, the dogs showed a preference for retrieving the neutral object, which was actually less salient than the alternative one. Dogs’ performance in this group indicates that they do not base their choices (solely) on the salience of the demonstration (although we do not exclude that this factor can play a part in the dogs’ choice).
The preference for retrieving the object with the more positive emotional display cannot be the result of the dogs’ ability to sniff out the object containing food, since in the Neutral–Neutral (control) group (the same two baits, both paired with neutral expressions), both the dogs’ first approach and their success in retrieving the baited bottle were at chance level. Moreover, similar to Schmidjell et al. (2012) and Hegedüs et al. (2013), we also did not find significant Clever Hans effect. Random performance was found in both variables in the control group designed to investigate the potential effect of owners’ voluntary/involuntary cues while the dogs were selecting an object. We also investigated whether the dogs’ performance in any of the conditions was influenced by simpler effects like preference for one side or for the object manipulated last by the owner, but none of these factors was found to have a significant effect on either of the measured variables. Learning during the experimental trials also did not explain the dogs’ performance, because repetition had no significant effect on either of the variables. Moreover, dogs already preferentially retrieved the bottle eliciting the more positive emotional display from the owner in the first trial in all except the Happy–Disgusted condition (which was only nearly significant). In the latter group, dogs were exposed to two highly salient emotional expressions, and their random choice in the first trial could indicates some limitation of their capacity to attend to these emotional messages within a short time frame (e.g. Range et al. 2009).
As a limitation of the study, we should mention that for some dogs, the fetching of the objects itself could provide a greater reward than food that makes the human emotional displays less relevant for making their choice (for parallel findings, see Sümegi et al. 2014). Since no reliable means was found to exclude the extremely motivated subjects, their performance might have biased our results.
In sum, we demonstrated that dogs are able to recognize the human emotional expressions of happiness and disgust. Their interest towards a ‘disgusting’ object may influence which object they approach first, but dogs are able to control their own preference and retrieve the object which is marked by the relatively more positive emotion of the owner. Based on these results, we conclude that both positive and negative emotions guide dogs’ behaviour in a two-object choice situation. Dogs demonstrate a preference for positive human emotions while also show avoidance of the negative ones.
This work was supported by the Hungarian Scientific Research Fund (K 84036), the Bolyai Foundation of the Hungarian Academy of Sciences, the MTA-ELTE Comparative Ethology Research Group (01 031), and the ESF Research Networking Programme ‘CompCog’: The Evolution of Social Cognition (www.compcog.org) (06-RNP-020). The authors are grateful to József Topál for his help in the development of the protocol. We would like to thank all the owners and dogs who participated in this study. We also would like to thank the three anonymous reviewers for all their useful comments on an earlier version of this paper.
Conflict of interest
The authors declare that they have no conflict of interest.
The experiment complies with the current laws of Hungary. According to the corresponding definition by law (‘1998. évi XXVIII. Törvény’ 3. §/9.—the Animal Protection Act), non-invasive studies on dogs are currently allowed to be done without any special permission in Hungary.