Inference in relational reasoning: A case study of Relational-Match-to-Sample
Kroupin, Ivan Georgievich
MetadataShow full item record
CitationKroupin, Ivan Georgievich. 2021. Inference in relational reasoning: A case study of Relational-Match-to-Sample. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.
AbstractAdult humans are uniquely proficient in reasoning with abstract relations (relational reasoning) - a capacity which underpins much of human-unique cognition, including scientific analogies, artistic metaphors and many phrases in day-to-day language and thought. An important question for cognitive science, therefore, is exploring the nature of this uniqueness - that is identifying the nature of population differences in relational reasoning between older children and adults, on the one hand and younger children and non-human animals on the other. The classic paradigm used to test for even the most rudimentary capacity for relational reasoning is Relational-Match-to-Sample (RMTS, Premack, 1983). RMTS involves matching pairs of objects on the basis of the relations same and different. As such, success on RMTS is evidence not only of a rudimentary capacity for relational reasoning, but also of representations of sameness and difference in unitary, symbolic format. If these representations are available early in life and/or across species, this would suggest that they may constitute primitives in an innate language of thought which underpins the combinatorial capacities of human language (and potentially non-human thought as well). Previous work has shown that non-human animals and children under the age of five struggle with RMTS, succeeding only after paradigms which may have led them to generate same/different representations for the first time. If this is the case, these populations are simply not able to engage in relational reasoning without special training. In contrast, a successful training paradigm used with crows and parrots provides a model of training (using simple Match-to-Sample, MTS tasks) which cannot have produced new relational representations. The effects of the original training were not, however, conclusive since other training mechanisms were also used with birds and these latter may have led them to generate same/different representations. The first paper demonstrates that MTS training can increase relational responding in human adults by changing inductive biases alone (since adults already have all of the necessary representations and capacities). This sets up the hypothesis that population differences in RMTS performance may sometimes not be differences in the availability in same/different representations, but rather differences in inductive biases alone, i.e. what bases of matching are inferred to be relevant - and that MTS training can bridge at least some population differences by changing these inductive biases alone. The second paper tests and confirms this hypothesis with four-year-old children, further predicting that children’s failure on RMTS are in some cases due to inductive biases leading them to specifically seek partial shape and/or color matches. The third paper tests and confirms this final predictions by making shape and color unlikely bases of matching by holding them constant across pairs in RMTS, which results in unprecedented wholly spontaneous success by four-year-olds on the task. In sum, these results demonstrate that population differences in relational reasoning performance, and RMTS specifically, are sometimes differences in inductive biases alone. This implies that relational reasoning and same/different representations may be available earlier in ontogeny and phylogeny than previously thought - and highlights the crucial role of inductive biases in the development and deployment of relational reasoning capacities.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37368317
- FAS Theses and Dissertations