Effects of Working Memory Load and Speaker Reliability on Contrastive Inference and Quantifier Processing

Abstract

This dissertation investigates the processing of two distinct linguistic phenomena: Quantifier interpretation and contrastive inference. Research has indicated that non-literal, or pragmatic interpretation, e.g., scalar implicature, is slower and more effortful than truth-conditional, or semantic, interpretation alone. Consequently, much research has been guided by the assumption that easy or fast processes are likely to be semantic in nature, whereas slow or effortful ones are likely to be pragmatic. In two eye-tracking experiments investigating the interpretation of scalar quantifiers and numerals, literal interpretation of "all" was impaired by verbal working memory load, violating the broad generalization that semantics is easy and pragmatics is hard. Exact interpretation of numerals was unaffected, suggesting numeral upper bounds are easy to compute. In two further experiments, contrastive inferences were easy to generate but difficult to suppress, a result not predicted by theories in which listeners generate such inferences by modeling speakers as rational agents. The effects of working memory load and speaker reliability on adjective interpretation confirmed instead the predictions of automatic egocentric processing. Taken together, these results challenge the division of linguistic processing into two homogeneous classes distinguishable by differences in cognitive implementation.