An adaptation of the goggles experiment for dogs: Evidence that dogs are unable to use their own experience to infer a human’s visual perspective

doi:10.21203/rs.3.rs-3787537/v1

Download PDF

Research Article

An adaptation of the goggles experiment for dogs: Evidence that dogs are unable to use their own experience to infer a human’s visual perspective

https://doi.org/10.21203/rs.3.rs-3787537/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Dogs have strong bonds with humans but the extent to which they can take our perspective into account is still in question. This study adapts the goggles experiment methodology to investigate if domestic dogs engage in level 2 perspective taking. The dogs not only have to understand when the human individual’s line of sight is blocked (level 1 perspective taking), but have to infer the other individual’s visual experience based on their own previous experience. A competitive context of food stealing is used.

Dogs (N = 24) first gained experience with two compartments, learning that food was accessible from one end while the other end had either a transparent or opaque lid. When stealing the food, dogs were visible to the human only through the transparent lid. In the test, dogs experienced two conditions: where both compartments were baited and a human either remained in the room (social condition) or left the room (non-social condition). Three factors were observed: (i) whether the dog took the food, (ii) the compartment the dog stole from, and (iii) latency to steal. The findings revealed that dogs didn't exhibit a preference for stealing from the concealed side, demonstrating that dogs don't consider the human's perspective when selecting which food to steal. However, in the social condition more dogs refrained from stealing, demonstrating sensitivity to the scenario’s competitive nature. It’s possible that the setting was overall too challenging and future research should look for alternative procedures to address the question of dogs’ understanding of a human’s perspective.

Domestic dogs

social cognition

visual perspective taking

goggles experiment

Recent studies have shown that domestic dogs might have some understanding of other individuals’ psychological states (Krupenye and Call 2019). This capability would be essential for the presence of Theory of Mind (ToM), which is the ability to ascribe mental states to others even if they differ from your own (Premack and Woodruff 1978). There is evidence that dogs are sensitive to others attentional state in communicative interactions (Soproni et al. 2002; Lakatos et al. 2012) and can differentiate whether someone is looking at them or away (Virányi et al. 2004) as well as being able to differentiate whether another individuals’ eyes are open or closed (Call et al. 2003). Dogs seem to also have some understanding of what Flavell et al.(1981) defined as Level 1 perspective taking (Bräuer et al. 2004), where an individual can use the line of sight of another individual to infer their visual perception. There is some evidence that dogs understand when another individual’s line of sight is blocked by a barrier (Kaminski et al. 2009) and are also sensitive to the level of light in a room when making the decision to steal (Kaminski et al. 2013). In the latter study the researchers set up a situation during which forbidden food was placed on the floor and the way the room was lit systematically varied. Either the face of the human, the piece of food or the face and the food were lit while the rest of the room was in darkness. The dogs’ behaviour indicated that they took the lighting in the room into account when making the decision to steal the food. They stole the food faster when the food was in darkness compared to when it was lit, while the lighting around the face of the human did not affect the dog’s behaviour indicating that dogs were not just guided by a simple associative rule of seeing the human’s eyes as an aversive stimulus (Neilands et al. 2020).

There is recent evidence suggesting that dogs understanding of others’ perspective might go beyond Level 1 perspective taking. For Level 2 perspective taking, an individual uses their own previous experience and mentally adopts the other individuals visual perception (Flavell et al. 1981). Maginnity & Grace (2014) presented dogs with a series of cups. Food was hidden under one of the cups. While the dog did not witness under which cup the food was hidden, the dogs observed one person (the knower) watching the hiding of the food. A second person (the guesser) was standing next to the food but facing away remaining ignorant of the location of the food. Now both people, the knower and the guesser, simultaneously pointed to different cups. Dogs distinguished between the two persons and followed the pointing gesture of the knower significantly more often than the pointing gesture of the guesser, suggesting that the dogs took the past visual experience of both people into account (but see Kaminski et al. 2009 for negative results regarding knowledge attribution in dogs).

While taken together this evidence suggests that dogs have some understanding of another individual’s visual perspective, there is as of yet no evidence that dogs mentally adopt another individual’s point of view (defined by Flavell et al. 1981 as ‘Level 2 perspective taking’), which would be evidence for a true ‘ToM’. One paradigm that was proposed to explore this is the so-called ‘goggles-paradigm’ which was first proposed by Cecilia Heyes (1998). In the proposed paradigm an individual gets the opportunity to gain experience with two pair of goggles, which look identical from the outside apart from a different coloured frame, but one is opaque while the other is transparent. The question is then if individuals who had experience with the googles can infer whether another individual wearing the googles can see through them. Using a blindfold instead of googles, the paradigm has been used testing human infants’ Level 2 perspective taking abilities (Meltzoff and Brooks 2008). The infants would either gain experience of the properties of a blindfold as transparent (despite appearing opaque from a distance) or not have any experience. Then after watching an adult put on the blindfold researchers measure whether the infants would follow the gaze of the individual wearing the blindfold (assuming the individual could see), or not (assuming the individual couldn’t see). The 18-month-old children were shown to follow humans gaze significantly more when they had experience that the blindfold is transparent than without the experience, implying they correctly inferred the other individual’s perspective from their own experience.

Due to the simplicity and the non-verbal nature of this paradigm, it has also been adapted for use in animal research (Karg et al. 2015; Krupenye & Call 2019). Karg et al. (2015) adapted the goggles method for use with chimpanzees in both a gaze following task, using hand held masks as the ‘goggles’, and a competitive food choice task. For the competitive food choice/later task the chimpanzees had to choose between two food options from inside compartments that had lids of differing transparencies, only one of which the competing human could see through. For both of these tasks the chimpanzee needs to make an inference about the human’s perspective using their own experience of the mask/compartment lids as either transparent or opaque despite them looking identical when they make their choice. The results of this study and that of Kano et al. (2019) provide evidence that chimpanzees, like human infants, are able infer a human’s perspective based on their own experience.

In the current study, we adapted the goggles method to test dogs’ Level 2 perspective taking abilities in a food stealing paradigm. Food was hidden in two compartments which had two distinct ends. From one end both compartments looked identical, while from the other end the compartments were covered with either a transparent or an opaque lid. The dogs gained experience of the properties of these lids while they were allowed to explore both ends of the compartments. Then we tested to what extend they used their knowledge to infer the visual perception of a human.

Subjects

24 domestic dogs (12 male, 12 female) of various breeds and aged between 1 and 10 years old (m = 5.75, sd = 3.10) participated in this study. An additional 17 dogs were excluded from the study as they were unable to meet the pre-test criteria. To take part the dogs needed to be medium to large size breeds due to the scale of the apparatus used. All subjects were living as pet dogs, and were recruited from a database of dog owners who volunteer their dogs to participate in research at the Dog Cognition Centre, Portsmouth, UK. The owners of the dogs were not present in the room during the study.

Apparatus

Testing took place in a large room (3.7m x 7.58m, see Fig. 1) at the Dog Cognition Centre, Portsmouth. The apparatus (see Fig. 2) comprised of one main board (170cm wide, 85cm high) with two compartments on one side. The compartments had a small opening on the back and at ground level which allowed the dogs to access the food placed inside. The compartments had liftable and interchangeable lids. Both lids were built with an identical wooden frame but either have an opening (transparent side) or were covered with a layer of fly screen and a layer of white backing fabric (opaque side). A handle at the bottom of the lid allowed for the experimenter to easily lift each lid, when the procedure required. Inside each compartment there was white flooring and three small lights, to ensure the food was clearly visible and had the same level of illumination. Additionally, on the lid side of the apparatus a stool was placed, 1.2m back from the main board and equidistance between the two compartments. Tape was used on the floor to mark the dogs’ starting position on the opening side of the apparatus.

Procedure

Throughout the procedure there is one experimenter and one dog handler, both female. The experimenter was the same individual for all participants, but the dog handler varied between dogs.

Pre-test

Every participant dog had to pass the criteria of a pre-test phase to take part in the study. To create the competitive context of the study the leave command was used, therefore only dogs who could follow a ‘leave’ command (or an equivalent command used by the owner) could participate. To test this, in an empty room distinct from the testing room, a piece of food was placed on the floor and the command ‘leave’ was given to the dog. If the dog tried to take the treat, the experimenter would cover the treat with their hand and repeat the command until the dog was able to meet the criteria of waiting, without trying to take the treat, for 30 seconds. If the dog was unable to meet this criterion after 10 minutes, they were excluded from the study.

To ensure the dog understood the competitive nature of the study, and the fact that a human could see them steal from the apparatus when no lids are present, there was a second stage to the pre-test. The participant dog was brought into the testing room, where the apparatus was present with no lids. The dog was led to the openings side and held by the dog handler. The experimenter entered and stood on the lid side of the apparatus, they gained the dog’s attention with a piece of food and then placed it in one compartment. From their side the dog could see it through the corresponding opening. The experimenter then gave the command ‘leave’ in a firm tone. If the dog started to approach the treat the experimenter again said leave and if they continued to approach, they stood up to assert authority, again stating the command. The dog needed to be able to wait 30s without taking the treat at this stage to be able to continue. This was then repeated in the other compartment.

If the dog was able to meet the pre-trial criteria, they were invited to take part in the familiarisation and test phase.

Familiarisation

The familiarisation phase provided the dog with experience of the properties of the lids i.e. transparent or opaque, as well as familiarisation of where they could access the food rewards after they were placed in the compartments. The dog handler led the dog into the room to the lid side of the apparatus, and clipped the dog’s lead onto a wall hook to position them at a distance to watch the handler. The handler showed the dog a treat, saying their name to gain their attention. Then they lifted a lid and placed the food in a compartment. The second lid was lifted to clearly show no treat being present in that compartment. The dog handler then led the dog around to the other side of the apparatus, to a predetermined point centred at equidistance from the openings. At this point the handler gave slack on the lead so the dog could then approach the apparatus and take the treat. If they approached the side without a treat, they were not allowed to then approach the side with the treat, but were guided back to the other side of the apparatus and the next trial started (alternating which side the treat was placed under). This provided the dogs with experience of making a choice between the two openings.

The familiarisation phase consisted of at least 10 repetitions and until the dog had reached the criteria of 5 consecutive correct trials (always choosing the side with the treat present), showing no hesitation taking treats from inside the opening. If they were unable to meet this criterion in 20 repetitions, they were not able to proceed to the testing phase.

Testing

There were two test conditions: Social and non-social.

Social condition

The dog handler followed similar methods to the familiarisation stage. They entered, attached the dog’s lead to the wall hook, and gained their attention with a treat. But unlike the familiarisation stage a treat was placed under both lids. The handler then walked the dog around to the predetermined marked position and at this point an experimenter entered the room and walked to the lid side of the apparatus. They then gained the dogs attention by saying their name, and gave a ‘leave’ command in a firm voice. The experimenter then sat on the stool (once seated their face remained in view of the dog), and focused their gaze at a spot on the main board of the apparatus between the two compartments. The handler then allowed slack on the lead and the dog was able to approach and steal one treat. The handler guided them away before they could try and take a second treat. The condition ended after 120 seconds irrespective of whether the dog did or did not make a choice (take a food item).

Non-social condition

The same procedure as the social condition was followed, however, once the leave command had been given rather than sitting on the stool, the experimenter walked out of the room, closing the door behind them.

The study had a within subject design where each participant dog took part in both of the conditions; social and non-social. Each participant took part in a total of 8 trials, 4 of each condition. To ensure the dog still followed the leave command, despite no repercussions from stealing during trials, the pre-test method in a separate room was repeated after every 2 trials. The order of the conditions was counterbalanced between dogs. The placement of the two lids (i.e. opaque on the left and transparent on the right-hand side) was alternated between participants.

Data analysis

The trials were video recorded and later coded for which compartment the dog stole food from (left or right hand side of the apparatus to check for a side bias), the lid of the compartment they stole from (opaque or transparent), and the latency to steal a piece of food. A secondary coder, unaware of the hypothesis coded 20% of all videos, and had strong intercoder reliability (Spearman = 0.88, n = 39 for Latency, and Kohens kappa = > 0.99, n = 77 for side and lid choice).

Ethical Approval

All the procedures were in accordance with the standards of the UK Animals (Scientific Procedures) Act 1986. The study received ethical approval from Animal Welfare and Ethical Review Body (AWERB), University of Portsmouth (Reference number 1221E).

To compare the stealing rate across conditions, the percentage of trials that the dog chose to steal a piece of food in each condition was calculated. A paired t-test was used to compare the percentages across conditions. Overall, stealing rates were significantly higher in the non-social condition compared to the social condition (t(23)=-1.90, p=0.06 (approaching the 95% significant level); Non-social; mean=91.7%, sd=22.9. Social; mean=84.4%, sd=33.6).

Looking at the latency to take a treat will allow us to investigate how hesitant each dog is to steal the food where a longer latency would indicate hesitancy to steal. A mean latency for each dog was calculated, only including trials where they chose to steal. A Wilcoxon signed-rank test compared the mean latencies per dog to steal a treat between conditions. The mean latency was higher for the social condition than the non-social condition (social; mean=11.6, sd=20.9, non-social; mean=7.45, sd=13.7) and the difference was shown to be statistically significant (z=-1.81, p=0.07, r=0.38).

In order to investigate if the dogs show a preference for the opaque lid compartment, the trials where the dog did not steal were removed and a percentage of the dog’s choice of the opaque lid was calculated for each condition (Figure 3). A paired t-test was used to compare the percentages of opaque lid choice in social and non-social conditions. There was no significant difference in lid choice between social and non-social conditions t(23)=-1.03, p=0.31.

To ensure there was no effect of learning on the dogs choices, the results of only the first trials of each dog were analysed. A fishers exact test was used to look for a preference to steal from under one of the lids in either condition. Results showed no overall preference is shown in the first trial (Fishers exact test p=0.70), which is consistent to the result shown when analysing all the trials.

To test if dogs showed a side bias, a paired t-test was used to compare the percentage dogs chose the left-hand side between conditions (Figure 3). There was a significant effect of condition where in the non-social condition there was a preference towards choosing the left hand compartment irrespective of the lid present, but no preference is seen in the social condition t(23)=2.01, p=0.05 (95% significant level) Social; mean=54.3, sd=46.4. Non-social; mean=71.9, sd=38.5.

In this study, we adapted the goggles paradigm for use with dogs. It is important to note that when the dog made a choice, both options looked identical to them and they had to use their own experience with the other end of the compartments to infer the perspective of the human and act accordingly. The results show that dogs did not differentiate between the compartments meaning that they did not take the human’s visual experience into account when making their decision to steal the food. It is possible that the setting was cognitively too challenging for the dogs, which might also explain why dogs developed a side preference. Research has shown that in situations which require a choice between two options, individuals sometimes develop a side bias if the task itself is cognitively too demanding (Gácsi et al. 2009).

Dogs were clearly sensitive to the competitive social context of the setting. Dogs stole significantly more during the non-social condition than they did during the social condition showing that they were sensitive to the context and are more hesitant to steal when a human was present. It is possible that in the current setting the overall requirements regarding dogs’ working memory were too high, which may overshadow dogs’ actual perspective taking abilities. This was proposed to explain dogs lack of perspective taking abilities in a study published by Bräuer et al. (2013). In that study forbidden food was placed in a tunnel through which the dogs had to put their paw to reach the food. One end of the tunnel was transparent, the other end opaque and dogs had to make the choice, which end to choose when reaching for the food. The human sat behind a barrier which had a small gap through which the human could see the dog’s paw enter the transparent, but not the opaque end of the tunnel. Dogs had the chance to briefly see the human sitting in position, but did not have visual access to the human while they approached either of the two tunnel ends. Dogs did not differentiate between the opaque and the transparent end when stealing the food, suggesting that they did not take the humans perspective into account. However, the setup required the dog to remember multiple locations (i.e. the treat, the tunnel it is hidden in, and the experimenter), and memorise them, while remembering how to access the food treat.

It is also possible that in the current study dogs might not have had sufficient experience with the properties of the lids. Their attention may have been focused more on the treat itself during the experience stage and the properties of the lid may not have been salient enough to draw their attention. However, we think that it is likely that the dogs had sufficient time in which to experience the properties of the lid, as the experience phase was kept as long as possible. It is also possible that the context may not have been motivating enough for the dogs, especially as there were no repercussions for stealing during the trials. The dogs might have quickly learnt they could steal at no cost. However, dogs did not differentiate between both compartments from the first trial on, suggesting that experience over time did not affect their behaviour.

We found an overall preference of the dogs during the non-social condition to steal food from the left compartment, irrespective of its properties. This could be due to the configuration of the room. The door from which the experimenter entered and left during the non-social condition, was located on the right-hand side of the room (closest to the compartment located on the right hand side). Therefore, when the human is leaving the room the compartment on the left is more concealed from the human’s view making it possibly more attractive for the dogs.

Overall, the results suggest that dogs’ cognitive abilities may be limited to first level perspective taking and paradigms requiring Level 2 perspective taking might be too demanding, which is consistent with the findings of Bräuer et al. (2013) and Kaminski et al. (2009). This may indicate that, while dogs seem to have some understanding of humans’ attention, gaze direction and line of sight, this might not be based on a true understanding of another individual’s point of view.

Competing interests

The authors have no competing interests to declare.

Funding

Funded by the University of Portsmouth.

Availability of data and material

No data associated in the manuscript.

Acknowledgements

We would like to thank all the dog owners who took the time to bring their dogs to participate. Thank you to Beth Madgwick, Hoi-Lam Jim and Hannah St John-Moss for assisting in data collection, and Nada Abbas for reliability coding.

Bräuer J, Call J, Tomasello M (2004) Visual perspective taking in dogs (Canis familiaris) in the presence of barriers. Appl Anim Behav Sci 88:299–317. https://doi.org/10.1016/j.applanim.2004.03.004
Bräuer J, Keckeisen M, Pitsch A, et al (2013) Domestic dogs conceal auditory but not visual information from others. Anim Cogn 16:351–359. https://doi.org/10.1007/s10071-012-0576-9
Call J, Bräuer J, Kaminski J, Tomasello M (2003) Domestic Dogs (Canis familiaris) Are Sensitive to the Attentional State of Humans. J Comp Psychol 117:257–263. https://doi.org/10.1037/0735-7036.117.3.257
Flavell JH, Everett BA, Croft K, Flavell ER (1981) Young Children’s Knowledge About Visual Perception: Further Evidence for the Level 1-Level 2 Distinction
Gácsi M, Kara E, Belényi B, et al (2009) The effect of development and individual differences in pointing comprehension of dogs. Anim Cogn 12:471–479. https://doi.org/10.1007/s10071-008-0208-6
Heyes CM (1998) Theory of mind in nonhuman primates. Behavioral and Brain Sciences 21:101–148
Kaminski J, Bräuer J, Call J, Tomasello M (2009) Domestic dogs are sensitive to a human’s perspective. Behaviour 146:979–998. https://doi.org/10.1163/156853908X395530
Kaminski J, Pitsch A, Tomasello M (2013) Dogs steal in the dark. Anim Cogn 16:385–394. https://doi.org/10.1007/s10071-012-0579-6
Karg K, Schmelz M, Call J, Tomasello M (2015) The goggles experiment: Can chimpanzees use self-experience to infer what a competitor can see? Anim Behav 105:211–221. https://doi.org/10.1016/j.anbehav.2015.04.028
Krupenye C, Call J (2019) Theory of mind in animals: Current and future directions. Wiley Interdiscip Rev Cogn Sci 10:1–25. https://doi.org/10.1002/wcs.1503
Lakatos G, Gácsi M, Topál J, Miklósi Á (2012) Comprehension and utilisation of pointing gestures and gazing in dog-human communication in relatively complex situations. Anim Cogn 15:201–213. https://doi.org/10.1007/s10071-011-0446-x
Maginnity ME, Grace RC (2014) Visual perspective taking by dogs (Canis familiaris) in a Guesser–Knower task: evidence for a canine theory of mind? Anim Cogn 17:1375–1392. https://doi.org/10.1007/s10071-014-0773-9
Meltzoff AN, Brooks R (2008) Self-Experience as a Mechanism for Learning About Others: A Training Study in Social Cognition. Dev Psychol 44:1257–1265. https://doi.org/10.1037/a0012888
Neilands P, Hassall R, Derks F, et al (2020) Watching eyes do not stop dogs stealing food: evidence against a general risk-aversion hypothesis for the watching-eye effect. Sci Rep 10:. https://doi.org/10.1038/s41598-020-58210-4
Premack D, Woodruff G (1978) Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences 1:515–526. https://doi.org/10.1017/S0140525X00076512
Soproni K, Miklósi Á, Topál J, Csányi V (2002) Dogs’ (Canis familiaris) responsiveness to human pointing gestures. J Comp Psychol 116:27–34. https://doi.org/10.1037/0735-7036.116.1.27
Virányi Z, Topál JÓ, Gácsi MÁ, et al (2004) Dogs respond appropriately to cues of humans’ attentional focus. Behavioural Processes 66:161–172. https://doi.org/10.1016/j.beproc.2004.01.012

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

An adaptation of the goggles experiment for dogs: Evidence that dogs are unable to use their own experience to infer a human’s visual perspective

Status:

Version 1

Abstract

Figures

Introduction

Methods

Pre-test

Familiarisation

Testing

Data analysis

Results

Discussion

Declarations

References

Additional Declarations

Status:

Version 1