Poster #39 - Overhypothesis Formation in Capuchin Monkeys (Sapajus sp.) and Children

• In Event: 2-163 - Poster Session 08
  In Poster Session: PS 08 Section - Cognitive Processes

Integrative Statement

The use of abstract higher level knowledge (e.g. overhypotheses) allows humans to learn quickly from sparse data, and make predictions in new situations. A study using Violation of Expectation methodology by Dewar & Xu (2010) suggests that this ability develops early in infancy, raising the possibility that it is part of our biological inheritance. To investigate the evolution of overhypothesis formation, we developed an action-based version of this task to test capuchin monkeys (Sapajus sp.) and 3- to 5-year-old human children. After seeing evidence (4 food items) sampled from three containers sorted either by item type or by item size, participants (monkeys: n=11, children: n = 80) were presented with two new test containers and one example item from each: a small, high-valued reward from A and a large, low-valued reward from B. Subsequently, subjects chose between two covert samples from these containers. If the participants learned the abstract rule governing the containers’ contents then they were expected to prefer the sample from A in the type condition to secure a high-quality item, but to prefer the sample from B in the size condition to secure a large item.
None of eleven capuchin monkeys showed this pattern, instead choosing indifferently in each condition (t-test, size: mean(s.e.)=51.50% (1.71%), t(10)=0.75, p=0.47; type: mean(s.e.)=51.88 % (2.54%), t(10)=0.89, p=0.40). However, as expected the children preferred the hidden item from the container with the small, high-value sample significantly above chance in the type condition (mean (s.e.)=71.25% (3.73%), t(39)=5.70, p<0.001). In contrast, they did not show a preference for either container in the size condition (mean (s.e.)=45% (3.44%), t(39)=1.45, p=0.15). In a follow-up task with reduced cognitive demands that focused on a simpler level of abstraction, performance of both species was very similar to their main experiment results. Here, subjects were presented with only two containers and four samples, respectively. The containers were either distinct in the food type (4 high-valued samples vs. 4 low-valued samples) or in the size of food items (4 large samples vs. 4 small samples). A previous preference testing ensured that all subjects preferred larger over smaller and high-valued over low-valued items. The children’s performance suggests a strong prior bias towards an organization by type or an enhanced focus on the type over the size or number of items.
Overall, these results hint towards a recent evolutionary emergence of this ability. Results of the capuchin monkeys and the children will be compared to a probabilistic hierarchical Bayesian model (Kemp, Perfors, & Tenenbaum, 2007) to examine in more detail how the species differ in forming general principles after observing a small amount of data. To disentangle the imposed cognitive demands by the sampling procedure and the species’ capacity for abstract knowledge formation, we are collecting data for a follow-up study that lets subject generate their own evidence for two conflicting overhypotheses.

Authors

• Elisa Felsche, University of St Andrews
  Presenting Author

• Patience Stevens, Carnegie Mellon University
  Non-Presenting Author

• Christoph Völter, University of St Andrews
  Non-Presenting Author

• Daphna Buchsbaum, University of Toronto
  Non-Presenting Author

• Amanda Seed, University of St Andrews
  Non-Presenting Author