Eye movements reveal fast, voice-specific priming

Abstract

In spoken word perception, voice specificity effects are well-documented: When people hear repeated words in some task, performance is generally better when repeated items are presented in their originally heard voices, relative to changed voices. A key theoretical question about voice specificity effects concerns their time-course: Some studies suggest that episodic traces exert their influence late in lexical processing (the time-course hypothesis; McLennan & Luce, 2005), whereas others suggest that episodic traces influence immediate, online processing. We report 2 eye-tracking studies investigating the time-course of voice-specific priming within and across cognitive tasks. In Experiment 1, participants performed modified lexical decision or semantic classification to words spoken by 4 speakers. The tasks required participants to click a red "x" or a blue "+" located randomly within separate visual half-fields, necessitating trial-by-trial visual search with consistent half-field response mapping. After a break, participants completed a second block with new and repeated items, half spoken in changed voices. Voice effects were robust very early, appearing in saccade initiation times. Experiment 2 replicated this pattern while changing tasks across blocks, ruling out a response priming account. In the General Discussion, we address the time-course hypothesis, focusing on the challenge it presents for empirical disconfirmation, and highlighting the broad importance of indexical effects, beyond studies of priming.

Figure 1

Schematic trial outline. Each trial began with a gaze-contingent 2000-ms fixation cross, followed by the onset of a spoken word for either lexical decision (LD) or semantic classification (SC). ‘Word’ (‘larger than a toaster’) decisions were made by locating and clicking a blue ‘+’; ‘nonword’ (‘smaller than a toaster’) decisions were made by clicking a red ‘×’. Response options were randomly located within the same visual half-field throughout the experiment, but changed locations in every trial. The dashed box in the center did not appear in the procedure, but is shown to illustrate a buffer zone where response icons could not appear.

Figure 2

Simulated sampling of “same-voice” and “different-voice” RTs, showing the relationship that emerges between item RT and voice effects. Panels A, C, E, and G show RT frequency distributions for hypothetical SV and DV words, sampled from flat distributions with equal variance, flat distributions with unequal variance, Gaussian distributions with unequal variance, and Weibull distributions with unequal variance (respectively). Panels B, D, F, and H show the associations that emerge between mean item RTs (SV+DV/2) and the size of voice effects (DV-SV).

Figure 3

Simulated sampling of “same-voice” and “different-voice” RTs, with the extra assumption that both versions of each word are loosely related to each other. The upper panel shows RT frequency distributions for hypothetical SV and DV words: SV words were sampled from a Gaussian distribution, and DV versions were created by sampling normally-distributed changes to those base RTs. The lower panel shows the association that emerges between mean item RTs (SV+DV/2) and the size of voice effects (DV-SV).