Nature: Scientific Reports wants you to know why your cat is making faces at you. They’ve published a study that uses artificial intelligence to decipher “CatFACS codings” — or, the expressions that cats make according “Facial Action Coding System” — in order to figure out how cats make faces to communicate with their owners:
Utilizing machine learning techniques and computational approaches can significantly enhance animal facial analysis. Automated tools have many advantages over manual coding, and they also have the potential to reduce subjectivity and bias in some cases53,54. Unlike manual methods, these tools apply general learned rules, which makes them applicable in different contexts. Automated facial signal coding has matured in the human domain, with numerous commercial tools available, such as FaceReader38and more. In animals, such approaches are just beginning to emerge, both in the context of grimace scale automation for pain detection55,56,57and for various emotional states classification58,59. One of the limitations of an automated approach is that automated tools require diverse training data, which usually implies careful preparation and a lot of preliminary work.
While developing animal analogs to FaceReader might require considerable resources (due to the data collection and validation), further interesting venues exist to explore. First, when AnimalFACS manual coding is already available, applying machine learning techniques can provide additional value, expanding and complementing the most commonly used statistical analysis methods. For instance, using the DogFACS coding, Boneh-Shirtrit59presented a decision trees-based model for classifying positive and negative emotional states in dogs. This approach showed lower classification performance than deep learning for the same task but was explainable and complemented findings discovered by performing statistical analysis related to emotional indicators. Secondly, the methods of geometric morphometrics present an alternative promising avenue for automated movement analysis. The tools for automated detection of facial landmarks were recently developed for domestic cats57,60. Using landmarks detected in the frames of the same video, it is possible to track the changes in an animal’s face through time and identify even the subtle facial movements, which is not possible with manual annotation61,62. This approach negates the issue with manual landmark annotation time cost, allowing for automated video processing and introducing the temporal dimension into the landmark-based methods for animal affective computing63.
The current study leverages these computational directions for investigating intraspecific social interactions in cats using the dataset of Scott and Florkiewicz1. First, we investigate to what extent machine learning algorithms can classify the context (affiliative/non-affiliative) of these interactions and what cat facial movements are more informative for this task. In this context, we specifically consider manual CatFACS-coded variables, studying the importance of the temporal dimension, i.e., the order in which the action units appear. Such an approach may demonstrate that cats interact by specific facial movements not only by “presenting” them but also by doing that in a specific order, and this order is important in terms of the context of interaction. We additionally apply the facial landmark detector as an alternative approach based on geometric morphometrics, providing a fully automated end-to-end pipeline to classify positive/negative interactions based on cat facial landmarks. The second contribution of this study is investigating whether domesticated cats exhibit facial mimicry. To our knowledge, this study is the first to address rapid facial mimicry in domesticated cats. As cats are known for their social flexibility, which enables them to coexist peacefully in multi-cat households and large colonies64,65,66,67,68,69,70,71, we predict that rapid facial mimicry would be more common in affiliative social interactions among cats, given its association with social bonding activities (such as play) in other mammals.