Understanding and remembering pragmatic inferences

Abstract

This dissertation examines the extent to which sentence interpretations are incrementally encoded in memory. While traditional models of sentence processing assume that comprehension results in a single interpretation, evidence from syntactic parsing indicates that initial misinterpretations are sometimes maintained in memory along with their revised counterparts (e.g., Christianson, Hollingworth, Halliwell & Ferreira, 2001). However, this evidence has largely come from experiments featuring sentences that are presented in isolation and words that are biased toward incorrect syntactic analyses. Because there is typically enough sentential context in natural speech to avoid the incorrect analysis (Roland, Elman, & Ferreira, 2006), it is unclear whether initial interpretations are incrementally encoded in memory when there is sufficient context. The scalar term “some” provides a test case where context is necessary to select between two interpretations, one based on semantics (some and possibly all) and one based on pragmatic inference (some but not all) (Horn, 1989). Although listeners strongly prefer the pragmatic interpretation (e.g., Van Tiel, Van Miltenburg, Zevakhina, & Geurts, 2016), prior research suggests that the semantic meaning is considered before the inference is adopted (Rips, 1975; Noveck & Posada, 2003; Bott & Noveck, 2004; Breheny, Katsos, & Williams, 2006; De Neys & Schaeken, 2007; Huang & Snedeker, 2009, 2011). I used a word-learning and recall task to show that there is evidence of the semantic meaning in the memory representation of sentences featuring “some,” even when the pragmatic interpretation is ultimately adopted. This raises two possibilities: first, the memory representation was of poor quality because both interpretations were available during encoding, or the semantic meaning was computed and encoded first and lingered even after the pragmatic interpretation was computed and encoded. Data from a conflict-adaptation experiment revealed a facilitating effect of cognitive control engagement. However, there was still a delay before the pragmatic inference was adopted. This suggests that only the semantic meaning is available initially and the system failed to override it in memory when the pragmatic interpretation was computed. Taken together, these findings demonstrate the incrementality of memory encoding during sentence processing.