Scales and scalarity: Processing scalar inferences

Abstract

Abstract The scalar word ‘some’ may be interpreted with an upper bound, i.e., as excluding ‘all’. Several studies have found that the computation of this scalar inference may be associated with a processing cost (e.g., Bott & Noveck, 2004; De Neys & Schaeken, 2007), which seems to argue in favour of theories according to which pragmatic inferencing is cognitively demanding (e.g., Sperber & Wilson, 1986). This argument holds on the premise that findings for ‘some’ can be generalised across the entire family of scalar words, which has been called into question by recent work highlighting the diversity within the class of scalar words (e.g., van Tiel, van Miltenburg, Zevakhina, & Geurts, 2016). In order to determine how generalisable the findings for ‘some’ are, we conducted three experiments in which we investigated the cognitive processing of seven scalar words that differ, inter alia, in their scalarity, i.e., whether they impose a lower (‘some’, ‘or’, ‘might’, ‘most’, ‘try’) or upper (‘low’, ‘scarce’) bound on their dimension. We find that the scalar inferences of the negatively scalar words ‘low’ and ‘scarce’ are not associated with a processing cost, unlike the scalar inferences of positively scalar words. We argue that the reported processing cost for scalar inferencing reflects increased cognitive demands associated with the processing of negative information.