Supervised versus Unsupervised Categorization: Two Sides of the Same Coin?

Abstract

This is the accepted version of the paper.This version of the publication may differ from the final published version. Permanent

Multiple feature use in pigeons’ category discrimination: The influence of stimulus set structure and the salience of stimulus differences.

“…Both models are based on the same idea, comparing within-category similarity to between-category similarity, with 'better' (easier to learn and more intuitive, cf. Pothos, Edwards & Perlman, 2011a) …”

Novel Labels Increase Category Coherence, But Only When People Have the Goal to Coordinate

“…But instead we found that these contextual factors modulated the categories that people used, and their social coherence. Thus, while our findings cannot tell us precisely which dimensions people selected for sorting, they do suggest that the dimensions that people choose during unsupervised categorization are somewhat flexible, in ways that are by definition independent of error‐based feedback (Goldstone, ; Pothos, Edwards, et al, ), and conversely are at least partially dependent on factors such as language and the context (or goal) of sorting. We note however that our participants were limited to categorizing the items into two groups only (i.e., rather than choosing their own number of categories; Pothos, Edwards, et al, ).…”

“…Thus, while our findings cannot tell us precisely which dimensions people selected for sorting, they do suggest that the dimensions that people choose during unsupervised categorization are somewhat flexible, in ways that are by definition independent of error‐based feedback (Goldstone, ; Pothos, Edwards, et al, ), and conversely are at least partially dependent on factors such as language and the context (or goal) of sorting. We note however that our participants were limited to categorizing the items into two groups only (i.e., rather than choosing their own number of categories; Pothos, Edwards, et al, ). In future work, it would be beneficial to examine the effects of novel labels alongside contextual factors (such as labels and the context of sorting) on category flexibility and social coherence when sorters are free to determine their own number of categories.…”

“…To an extent, language relies upon this category coherence, because people need to have a similar understanding of the world—and how objects within it are labeled—in order to successfully communicate about the world. Children presumably learn this shared understanding through the feedback that they receive as they acquire language; for example if a child incorrectly labels a lemon as an orange, an adult can produce the correct label, “lemon,” and so provide feedback about the underlying conventional category and its associated label, in a process of “supervised categorization” (Pothos, Edwards, & Perlman, ).…”

“…However, the majority of work examining the effects of novel labels on adult categorization (e.g., Barnhart et al, ; Lupyan & Casasanto, ; Lupyan et al, ) involves supervised categorization paradigms, in which people categorize objects using externally specified dimensions and receive feedback as they learn to do so. Pothos, Edwards, et al () argued that this feedback promotes representational flexibility in how people choose their dimensions for categorization; that is, the way in which people categorize can be changed over time through error‐based feedback (Goldstone, ). In contrast, unsupervised categorization (e.g., Malt et al, ) is spontaneous and can be unpredictable in how people choose dimensions for categorization (Pothos & Chater, ).…”