Towards testing of identity concepts in horses: conditional discrimination learning∗

Despite a vast amount of literature about horse training, there is little scientifically verified knowledge about learning behavior in horses. See Murphy and Arkins () for a review of research about learning behavior in horses. In particular, there is little research about horses’ abilities of concept formation and relational learning. However, some researchers have found evidence for learning processes that can be seen as indicators of relatively advanced inference abilities in horses. Hanggi () demonstrated that horses can learn to select symbols with an open center in two-choice discrimination tasks where one stimulus had an open center and the other had a filled center. After the horses had learned this task for a small set of stimuli, they carried out the same task for novel stimuli with a performance that was significantly above chance. Hanggi () investigated discrimination based on relative size and showed that this ability was transferred to novel symbols and from symbols to solid objects. Furthermore, the investigations by Roth et al. () show that horses can select colored stimuli based on relative hue. Flannery () addressed whether horses can use a concept of visual identity. However, the task that the horses learned can be carried out with a strategy based on direct associations between symbols and rewards. Thus, the results presented by Flannery () can be explained without the usage of a general concept of visual similarity. In this thesis, I tested to what extent horses learned a matching-to-sample task, and I also tested the learning performance of the horses in two other tasks that are