New paradigms to study numerical cognition in animals

The mathematical abilities of animals have long fascinated scientists and the general public alike. However, it is only in the last two decades that researchers have systematically investigated this issue. Numerical abilities are widespread among mammals, birds and fish but little is known about their occurrence in other taxa. They play important functions in several ecological contexts, such as foraging, mate choice, and social interaction. The actual mechanisms underlying number sense in animals and the upper limits of these abilities are still a matter of debate. Conflicting results often emerge when comparing evidence from different species and studies of the same species. It is unclear to what extent this is due to differences in the complexity of the nervous system, to ecology, to phylogeny, or to other factors. Some have argued that intra- and inter-species differences may be due to differences in the used methodology. This is frequent in human literature, but the issue has received little attention in animals. To unravel these issues, we need to expand the number of studied species by including species that belong to a less studied taxa. In addition, there is a need for studies that directly compare different methodologies within the same laboratory. In this thesis, I focus on the two latter issues. In two studies, I investigated the effect of the used methodology on the assessment of numerical abilities in a small teleost fish, Poecilia reticulata. In the first study, I adopted an operant conditioning procedure to train fish to solve a numerical discrimination task using computer-generated stimuli. With this procedure, fish showed very low numerical discrimination performance, considerably lower than that reported in previous studies carried out in the same laboratory, in which fish had to solve a similar numerical task in a more naturalistic setting. In a series of subsequent experiments, I attempted to identify the key factors that determine the different outcome of the two methods. Results only partly answer the question and it is possible that there are non-cognitive factors (i.e. side bias due to cerebral lateralization) that explain the observed difference. Previous studies on guppies showed greater numerical acuity when tested with operant conditioning procedures than when tested with spontaneous preference methods, such as testing the preference for the larger of two social groups (shoal choice test). This difference could be due either to the different accuracy of the two methodologies or to within-species variation in acuity related to the examined context (foraging vs social). In the second study, I set up a modified version of the shoal choice test in order to eliminate some of the limitations of the method that are thought to determine a large imprecision in the measurement. With this new method, guppies were able to discriminate 4 vs. 5 companions, the same limit of numerical discrimination determined with operant conditioning procedures. The last two studies focused on studying numerical cognition in two new species, an amphibian (Hyla intermedia) and an arthropod (Acheta domesticus) in a new context, shelter choice. Both species showed a preference for the larger cluster of shelters and, in a series of subsequent experiments, I tried to determine the mechanism involved in such discrimination. H. intermedia appears to possess a genuine numerical system, being able to select the most numerous set even after I experimentally controlled the perceptual continuous variables, such as cumulative surface area and convex hull, that covary with number. Results suggest that A. domesticus have a less sophisticated system