Deciding to Look: Revisiting the Link between Lexical Activations and Eye Movements in the Visual World Paradigm in Japanese

All current theories of spoken word recognition (e.g., Allopenna et al., 1998; McClelland & Elman, 1986; Norris, 1994) suggest that any part of a target word triggers activation of candidate words. Visual world paradigm studies have relied on the linking hypothesis that the probability of looking at the referent of a word directly tracks the word’s level of activation (e.g., Allopenna et al., 1998). However, how much information is needed to trigger a saccade to a visual representation of the word’s referent? To address this question, the present study manipulated the number and location of shared segments between the target and competitor words. Experimental evidence is provided by two visual world paradigm experiments on Japanese, using natural and synthesized speech. In both experiments, cohort competitor pictures were not fixated more than unrelated distractor pictures unless the cohort competitor shares the initial CVC with the target. Bayesian analyses provide strong support for the null hypothesis that shorter overlap does not affect eye movements. The results suggest that a listener needs to accumulate enough evidence for a word before a saccade is generated. The human data were validated by an interactive computational model (TRACE: McClelland & Elman, 1986). The model was adapted to Japanese language to examine whether the TRACE model predicts competitor effects that fit human data. The model predicted that there should be effects when words share any amount with a target which confirms the current theory. However, the model did not fit the human data unless there is longer overlap between words. This indicates that eye movements are not as closely tied to fixation probabilities of lexical representations as previously believed. The present study suggests that looking at a referent of a word is a decision, made when the word’s activation exceeds a context-specific threshold. Subthreshold activations do not drive saccades. The present study conclude that decision-making processes need to be incorporated in models linking word activation to eye movements