Abstract: Two baboons (Papio papio) successfully learned relational matching-to-sample: They picked the choice display that involved the same relation among 16 pictures (same or different) as the sample display, although the sample display shared no pictures with the choice displays. The baboons generalized relational matching behavior to sample displays created from novel pictures. Further experiments varying the number of sample pictures and the mixture of same and different sample pictures suggested that entropy plays a key role in the baboons' conceptual behavior. Two humans (Homo sapiens) were similarly trained and tested; their behavior was both similar to and different from the baboons' behavior. The results suggest that animals other than humans and chimpanzees can discriminate the relation between relations. They further suggest that entropy detection may underlie same-different conceptualization, but that additional processes may participate in human conceptualization.

Abstract: This article reviews the prerequisites for quantitative models of animal learning and cognition, describes the types of models, provides a rationale for the development of such quantitative models, describes criteria for their evaluation, and makes recommendations for the next generation of quantitative models. A modular approach to the development of models is described in which a procedure is considered as a generator of stimuli and a model is considered as a generator of responses. The goal is to develop models that, in combination with many different procedures, produce sequences of times of occurrence of events (stimuli and responses) that are indistinguishable from those produced by the animal under many experimental procedures and data analysis techniques.

Abstract: Traditionally, dogs have been trained by operant conditioning techniques; that is, dogs make a desired behavioural response and this response is reinforced by a reward such as food. This type of training is very effective in training dogs to perform basic obedience behaviours (e.g. `stay'). However, dogs are social animals and should be predisposed to learn from social stimuli. In the present study, we used a modified version of the model-rival technique that has been extensively used in experiments investigating the cognitive ability of parrots. In this technique, social stimuli are used to create in the animal an interest in the object without the use of food or other rewards. Therefore, the animal learns the name of the object (intrinsic reward) and not that the object's name means food. In this experiment we compared the learning ability of nine pet dogs to solve the same retrieval-selection task having been previously trained using operant conditioning or model-rival techniques. The retrieval-selection task was the dogs had to correctly select the commanded object to bring to the experimenter from a group of three similar objects. The results show no difference in the speeds with which the dogs solved the test--demonstrating the efficacy of the model-rival method. This is the first time that the effectiveness of the model-rival technique has been experimentally demonstrated with dogs. Furthermore, we believe that the methodology reported in this paper has applications in dog training and in experiments into dog cognition.

Abstract: Sixteen horses were used to determine if number of trials given per training session (5, 10, 15 or 20) affected learning performance in an avoidance conditioning task. The horse had to move from one side of a test pen to the other during an auditory cue presentation to avoid aversive stimulation. A pen 8 mx3.6 m, divided into two equal sections by a 13-cm diameter plastic pipe lying on the ground, was used as the test pen. Painted plywood panels were fastened to the fence in half the pen to help horses distinguish visually between the two parts. A 10-s auditory cue was used as a signal for horses to move from one side of the test pen to the other. A 20-s intertrial interval was used. Training sessions were conducted every third day. Each trial was recorded as an avoidance (the horse completed the task during auditory cue presentation and avoided aversive stimulus) or an error (the horse received aversive stimulus). After completing ten consecutive avoidances (criterion), the horse was removed from the study. Numbers of training sessions, trials, avoidances and errors until reaching criterion were recorded for each horse. Horses varied greatly within these variables with ranges of 3-18 sessions, 37-121 trials, 20-68 avoidances and 17-53 errors to criterion. No differences were detected (P>0.05) in the number of conditioning trials per training session (treatment) for the mean number of trials, avoidances or errors to criterion. Number of training sessions to criterion differed (P<0.01) among treatments, indicating that an optimum number of learning trials per training session might exist. Mean sessions to criterion for horses receiving 5, 10, 15 and 20 trials per session were 15.1+/-1.3, 5.8+/-1.1, 5.3+/-1.1 and 4.6+/-1.1, respectively. Regression analysis indicated that 16.2 trials per training session would minimize number of sessions to criterion. Although it is widely assumed that learning efficiency in horses is decreased when intense activity is concentrated into a small number of sessions, these results indicate that moderate repetition of training activities is needed for efficient learning.