Faces and Tails of Emotions: Using Citizen Science and Automated Methods to Assess Emotions in Dogs

Abstract

Emotion expressions are facial and bodily displays that co-occur with emotional response or elicitation and can o↵er a non-invasive and instantaneous assessment of emotions, pain, and welfare in animals. Objective and standardised measures of emotion expressions can be challenging, especially in species with large morphological di↵erences. Manual coding of facial and body expressions is prone to errors because of observer fatigue, inconsistencies in methodology and subtle movements may be missed. Machine learning methods that auto matically label and track the facial and body movements from videos may obviate human error and may be more consistent. Dogs navigate social interactions with multiple partners; us, other dogs, and other domes ticated animals. This leads to questions about how and by which modes they communi cate and express emotions. In this thesis I explore how and if dogs express emotions with facial expressions and tail movements and if an audience e↵ect influences tail behaviour. Through two citizen science paradigms, dog owners carried out emotion eliciting experi ments with their dogs while recording them. The first study only consisted of emotion elic itation whereas the second study included a communication component where dog own ers spoke with and ignored their dogs during emotion elicitation. I used semi-automated methods to label and extract facial and tail features and explored which features provided information about the emotion, and communication conditions, the dogs experienced by comparing the classification accuracy. I validated the methods through behavioural analy sis. Changes in behaviour were consistent with the expected arousal components of the emo tions elicited in the experiments. However, the dogs were more stressed than expected dur ing the condition where owners pet them. The automated labeling errors were comparable to manual labeling errors. Facial features provided information to accurately classify the emotion eliciting conditions the dogs experienced whereas the tail features only provided information to accurately classify the communication condition. Limitations and future di rections are discussed in the thesis as well as the advantages and disadvantages of a citizen science approach.