Abstract

Research on predictive processing has shown that listeners anticipate upcoming linguistic information using linguistic cues, such as verb meaning. Beyond information in the input, listeners’ prediction can also be influenced by top-down control, such as goals and strategies developed in response to a given task. However, it is unclear to what extent evaluations of task relevance modulate top-down control of prediction. We present evidence from two eye-tracking studies in the Visual World Paradigm that associative predictions are automatic and not influenced by implicit shifts in task relevance. In Experiment One, we presented 121 participants with 68 biclausal sentences containing a Male and Female referent per clause (e.g., After John answered his phone, Mary sighed and produced her wallet) while looking at diplays of three object-images, two of which corresponded with the objects heard in the sentence. Some of these sentences contained predictive verbs (e.g., answer) which constrained for the following target object; in the experimental items, these verbs always occurred in the first clause as the action of the Male referent. We also introduced a between-group attentional manipulation, with participants being instructed to pay attention to either gendered referent and answer comprehension questions. Since predictive cues in the experimental items were always associated with Male referent, we argued that prediction was task-relevant for Male-attending listeners, but not for Female-attending listeners. Thus, we expected that the Female-attending participants would display fewer and reduced predictive fixations to the target when compared with the Male-attending participants in the predictive window. But, in spite of the task-irrelevance of prediction for the Female-attending group, we failed to find an intergroup difference in either the extent or onset of prediction, suggesting that verb-based, associative predictions are not sufficiently demanding (and are likely automatic) and so do not provoke top-down strategizing w/r/t predictive processing.

However, when task relevance is made explicit in the form of an instruction to predict, prediction is upregulated irrespective of the verb and moreover initiated earlier, likely due to an enhancement of predictive visual search. We demonstrated this in Experiment Two, in which 60 new participants received the same materials as Experiment One but received instructions to actively predict the actions of the Male referent. With this new explicit task to predict, we expected participants to display earlier and more frequent anticipatory fixations relative to the participants of Experiment One, for whom the task relevance of prediction was implicit and not specified. In line with these predictions, Experiment Two participants dsiplayed earlier and more frequent predictive fixations, though this upregulation was irrespective of whether the first-clause verb was predictive or neutral. Task relevance is thus modulatory for associative prediction when it is specified for listeners.

We also conducted an exploratory analysis of the relation between self-reported confidence in one’s trial-by-trial predictions and predictive fixations. Participants in Experiment Two also received questions prompting them to report their confidence in each trial’s prediction and whether this prediction had been accurate. Broadly, we found convincing evidence that increased confidence is associated with increased prediction. However, there are important methodological (i.e., were participants influenced in their confidence reportage by the dis/confirmation of their prediction?) and theoretical (i.e., what exactly is confidence in this context?) questions which limit the generalizability and informativity of these data. For a more nuanced discussion, we report a full results set and analysis + discussion as part of our supplementary materials on OSF.

Current Status

Our draft manuscript has been editorially accepted to Language, Cognition, and Neuroscience – we now await peer review responses!

External Links

View this project’s preregistration, open-source materials, and analysis pipelines. For past presentational material (and a humbling vista over how our thinking has changed throughout this project), see here. To read the first draft of our manuscript (currently in prep), see our upload to PsyArXiv via OSF.