Adjectival vs. nominal categorization processes: The Rule vs. Similarity hypothesis

Classification of entities into categories can be determined based on a rule – a single criterion or relatively few criteria combined with logical operations like ‘and’ or ‘or’. Alternatively, classification can be based on similarity to prototypical examples, i.e. an overall degree of match to prototypical values on multiple dimensions. Two cognitive systems are reported in the literature to underlie processing by rules vs. similarity. This paper presents a novel thesis according to which adjectives and nouns trigger processing by the rule vs. similarity systems, respectively. The paper defends the thesis that nouns are conceptually gradable and multidimensional, but, unlike adjectives, their dimensions are integrated through similarity operations, like weighted sums, to yield an overall degree of match to ideal values on multiple dimensions. By contrast, adjectives are associated with single dimensions, or several dimensions bound by logical operations, such as ‘and’ and ‘or’. In accordance, nouns are predicted to differ from adjectives semantically, developmentally, and processing-wise. Similarity-based dimension integration is implicit – processing is automatic, fast, and beyond speaker awareness – whereas logical, rule-based dimension integration is explicit, and is acquired late. The paper highlights a number of links between findings reported in the literature about rule- vs. similarity-based categorization and corresponding structural, distributional, neural and developmental findings about adjectives and nouns. These links suggest that the rule vs. similarity (RS) hypothesis for the adjective-noun distinction should be studied more directly in the future.