You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster E27, Friday, November 10, 10:00 – 11:15 am, Harborview and Loch Raven Ballrooms

Making sense of real-time access to knowledge during sentence processing: What you know, what you don’t know, and what you don’t know you know

Melissa Troyer1, Marta Kutas1;1University of California, San Diego

A fundamental part of understanding language involves dynamically connecting linguistic input (e.g., words) with neural knowledge representations. A well-known event-related brain potential—the N400—provides a window into the neurocognitive access to meaning via verbal input. N400 amplitude is sensitive to semantic relationships between an incoming word and the ongoing context, with decreased N400s reflecting greater ease of semantic access. For example, a word’s predictability, operationalized as its offline cloze probability (i.e., the proportion of people who produce a specific word given a context), is strongly correlated with that word’s mean N400 amplitude. Typically, researchers correlate mean cloze probabilities (from one group of participants) with mean ERP measures (in a separate group). The precise nature of the relationship between offline cloze probability and online predictability, especially within a given individual, however, remains an open question: does N400 amplitude index (1) all-or-nothing (pre-)activation of a word in context, or as some have suggested (2) graded/partial (pre-)activation of a word in context as a function of an individual’s linguistic and world knowledge, among other factors? Because it is infeasible to estimate the entirety of someone’s world knowledge in a typical laboratory experiment, we used the narrative world of Harry Potter. Undergraduate students more or less knowledgeable about Harry Potter read sentence pairs about the Harry Potter domain while we recorded EEG/ERPs. The final word of each pair was perfectly predictable from the context, assuming perfect knowledge of the Harry Potter stories. After each sentence pair, participants were asked to indicate whether they had known the stated information when they first read it or not. At the end of the ERP recording, we independently assessed the participants’ knowledge of the Harry Potter domain via a trivia quiz (which determined their knowledge score, range: 11-38 questions out of 40). Unsurprisingly, high-knowledge individuals reported knowing more items than low-knowledge individuals. Also, as expected, across all participants, N400 amplitudes were dramatically reduced for items that individuals reported having known vs. not. More remarkably, knowledge scores were systematically related to N400 amplitudes, but this relationship was reliable only for items that participants reported not having known/remembered during the online task: compared to less-knowledgeable peers, individuals with greater knowledge exhibited smaller (more positive) N400 amplitudes for items they reported not knowing. One possible interpretation is that individuals with greater knowledge have a higher threshold for willingness to report what they know. We propose an alternative, albeit non-mutually-exclusive, interpretation: namely, that high-knowledge individuals may enjoy (or suffer from) access to broader and/or deeper semantic networks, implicitly accessing information related to the ongoing context in the absence of explicit recognition. Whatever the explanation, our results rule out accounts on which (pre-)activation of words in context is strictly all-or-nothing and suggest that by as early as 200 ms, fine-grained differences in the functional organization of individual-level knowledge may at least partially determine graded semantic activation during real-time sentence reading.

Topic Area: Meaning: Combinatorial Semantics

Back to Poster Schedule