Inferences about members of kinds: The generics hypothesis

Abstract

People routinely make inferences based on kind membership. For example, if you were told that a particular kind of animal is a tiger, then you would likely infer that it has stripes. Under what conditions are people willing to infer that a member of a given kind has a property? Two hypotheses were examined. The base rate or prevalence hypothesis holds that people rely only on their knowledge of the statistical frequency of a property among its kind to infer whether a member has that property. An alternative is the generics hypothesis, which states that people are influenced by their belief that the relevant generic generalization is true. In other words, if people agree to the generalization, “ducks lay eggs”, then they should be willing to make the inference that an arbitrary individual duck lays eggs, despite their knowledge that the majority of ducks do not lay eggs (i.e., juveniles, males, and infertile females). We present data that support the second hypothesis. Rather than being driven solely by beliefs about prevalence, agreement to the relevant generic predicted performance on an inference task beyond estimated prevalence or cue validity. These findings suggest that models of categorization that are based solely on statistical or simple probabilistic principles are incomplete. They also provide support for the idea that generics articulate core conceptual beliefs that guide our interactions with the world.

Keywords:

• Generics
• Generics hypothesis
• Default inferences
• Non-monotonic reasoning
• Concepts

Acknowledgements

This work was supported in part by a National Science Foundation Graduate Research Fellowship awarded to Sangeet Khemlani. We thank Jeremy Boyd, Adele Goldberg, Phil Johnson-Laird, Mark Johnston, Olivia Kang, Niklas Kunze, Max Lotstein, Adam Moore, and Lance Rips for their many helpful discussions and suggestions.

Notes

¹We also ran a separate, counterbalanced block of the generic agreement task on the participants in the main experiment. For brevity, these data were not used as predictors in any of the analyses in the paper, though they produced effects comparable to the generic agreement data collected in the norming study (as analyzed in Khemlani, Leslie, & Glucksberg, 2009). The participants’ default inferences were not affected by whether the generic agreement task or the inference task came first, and so the two samples were collapsed in all subsequent analyses.