Poster E13, Saturday, August 18, 3:00 – 4:45 pm, Room 2000AB

Electrophysiological Evidence of Early Lexical Influences on Sub-lexical Processing: Evidence from the Ganong Paradigm

Colin Noe1, Simon Fischer-Baum1;1Rice University, Department of Psychology

Theories of speech perception differ on whether there is only feedforward activation from sublexical levels of representation to lexical levels or whether top-down lexical and semantic information can influence sublexical processing. It has been well established that top-down influences bias what subjects report hearing (e.g. Ganong, 1980; Connine & Clifton 1987), but this bias could either reflect top-down processing or bias at a later stage, such as a response selection. Behavioral evidence alone cannot distinguish when exactly top-down influence is biasing perception. Thus, it cannot conclusively rule out response selection bias explanations of top-down influence. The current study looks at a classic effect of top-down influence on sub-lexical perception -- the Ganong lexical effect -- but includes an online measure of sub-lexical encoding, to verify the time-course of the top-down bias effect. Specifically, the current study compares how the N1 ERP waveform, which indexes the encoding of the sub-lexical feature voice onset time (VOT; Toscano et al., 2010), is influenced by Ganong lexical bias. Twenty-one participants participated in a Ganong two-alternative forced choice experiment while having EEG continuously recorded. Participants made stop-consonant voicing judgements on 9 step VOT continua varying between /d/ to /t/ and /g/ to /k/. The VOT continua were embedded into lexically biasing word environments (e.g. /d/-/t/ in dape-tape versus in date-tate). We test whether sublexical processing, as measured by the N1, is sensitive to both bottom-up information, here variations in VOT, and top-down information, here the bias in categorical perception created by the lexical context. Behaviorally, we replicated the Ganong effect, finding that participants were more likely to the familiar word percept with ambiguous VOTs. In terms of ERP, we replicated Toscano et al. (2010), finding a linear relationship between VOT and N1 amplitude. Critically, we observe an effect of lexical context on N1 amplitude, with N1 amplitude of ambiguous stimulus shifted in the direction of the lexically-favored endpoint. For instance, an ambiguous /d-t/ in dape-tape was encoded sub-lexically more like a /t/. Conversely, in date-tate, that same VOT-step was encoded more like a /d/. The main effects of VOT and bias were moderated by an interaction; the top-down influence of lexical context exerts a greater for the most ambiguous VOTs. The magnitude of the lexical bias effect changes over time, first reaching significant levels at a 25-75 msec time window, and then rapidly growing from 100-175 msec. The observation of an interaction of lexical and phonological information early after stimulus onset which grows over time and is sensitive to stimulus VOT information supports top-down feedback accounts of speech perception.

Topic Area: Perception: Speech Perception and Audiovisual Integration