Feature Integration as Prediction Coding: Evidence from a Prime-Probe Paradigm

The prevalent view of visual perception is that the world’s simple features are first encoded and then bound together into more complex environments. Action planning is thought to be achieved through the anticipation of desired perceptual effects, allowing action planning to interact with perception. Evidence supporting stimulus-response binding comes from the partial repetition cost paradigm wherein there is a cost for partially repeating a previous stimulus-response episode. Recently, however, a prime-probe paradigm has been introduced in which feature binding effects are absent despite conditions in which they should be present. This abnormally suggests that feature binding theories may need to be reconsidered. Here, I test the thesis that the partial repetition costs currently viewed as support for feature binding approaches to perception are better explained via prediction coding. To do so, I first show that partial repetition costs are absent from the prime-probe paradigm with the effects being purely attentional. Then, across five additional experiments, I systematically alter the prime-probe paradigm to determine the boundary of when partial repetitions will appear. In particular, I manipulated the prime and probe stimuli’s spatial overlap, the response set overlap, and whether the responses are spatial or not. I found that feature binding effects did not appear until there was spatial overlap, the response sets were the same, and the responses were both spatial. That is to say, that feature binding effects were absent until there was substantial overlap between the prime and probe tasks. I discuss these findings as support for the prediction code view of partial repetition costs.Ph.D