Relational Induction and Stimuli Dimension Effects in the Relational Matching Tasks

• In Event: 1-198 - Acquisition of Same/Different Relations by Human and Nonhuman Groups

Integrative Statement

Relational-Match-to-Sample (RMTS, Premack, 1983) is a standard test of relational reasoning abilities - specifically of reasoning with the relations same and different. Across many studies (e.g. Premack, 1983; Hochmann et al., 2017) non-human animals and children under five have been shown to fail RMTS in versions where both same and different are used as sample relations. This has lead some (e.g. Penn, Holyoak & Povinelli, 2008) to argue that young children and non-human animals have representations of same and different which differ qualitatively from those of older children and adults, such that the latter but not the former support RMTS success.

Work by Smirnova and colleagues (2015; 2015), however, has shown non-human animals (crows and parrots) are capable of spontaneous RMTS success given training on certain non-relational match-to-sample (MTS) tasks. In a series of studies, we show training on these very same MTS tasks also increases relational matching adults (in a task where matches can either be made based on object features or the relation same) and produces spontaneous success on RMTS in four-year-olds (a population which otherwise fails).

Given MTS training cannot plausibly change relational representations of children or non-human animals (much less adults), we suggest that their failure on RMTS is a result of not seeing the relations same and different as relevant bases of matching - despite having the requisite representations. The effect of MTS training, then, is changing how these populations solve the inductive problem of what bases of matching are relevant to the task such that they are more likely to see the relations same and different as relevant bases.

Exploring the inductive problem in RMTS leads us to consider other factors influencing whether the relations same and different are inferred as task-relevant. In a series of studies we explore one such factor: Adults were given a matching game where possible responses are either matching cards because they both contain an identical object or matching cards because both cards contain identical objects (i.e. matching on the relation same). When objects in the task vary in color and shape, 41% of adults consistently match on object identity and not relations. In stark contrast, when objects vary only in size, just 1% of adults consistently matched on object identity. This provides strong evidence for the dimension on which stimuli vary being an important factor in solving the inductive problem of when the representation same is seen as relevant.
In sum, this work supports the central importance of relational induction - identifying which relations are relevant in a given context - in relational reasoning generally and in studying representations of same and different specifically. First, without an account of the major factors which determine whether and when relational representations are actually applied to the world, we consistently run the risk of mistaking problems of induction for issues of representation. Second, such an account is central to understanding how we decide whether and when such representations are meaningful and thus how we use them to shape ourselves and the world around us.

Authors

• Ivan Kroupin, Harvard University
  Presenting Author

• Susan Carey, Harvard University
  Non-Presenting Author