Abstract
Previous work using simple stimuli argued that the P300 can serve as an index of visual similarity (e.g. Azizian, Freitas, Watson, & Squires, 2006), but this work did not rule out the possibility that semantic similarity masqueraded as visual similarity. Does the P300 provide a neural indicator of semantic similarity? To answer this question, we collected visual and semantic similarity ratings between a target category of objects (butterflies) and 2,000 random category nontargets. ERPs were recorded during a foveal discrimination task: Subjects identified via button press whether each of 950 images—displayed sequentially—was a butterfly (1/5[sup]th[/sup] of trials) or not. Nontarget items were selected to be either: semantically but not visually similar to butterflies, visually but not semantically similar to butterflies, dissimilar on both ratings, or similar on both. We found that P300s had larger amplitudes for semantically-similar and both-similar nontargets compared to visually-similar and both-dissimilar nontargets. In a control experiment, subjects were shown the identical images but were told to give the same speeded response to every object regardless of its category. Under these conditions, the P300 was unaffected by visual or semantic similarity, consistent with similarity being no longer task relevant. This demonstrates that the modulation of the P300 was due to semantic similarity between targets and nontargets, and not simple item effects. In conclusion, we demonstrated a clear relationship between the P300 and semantic similarity using realistic objects, but not visual similarity as had been previously suggested in the context of simpler stimuli. Our findings also extend work showing P300-like responses during viewing of scenes with semantic inconsistencies (Ganis & Kutas, 2003), suggesting that the P300s preceding N400s in scene viewing may reflect the establishment of semantic relationships and not merely infrequency of incongruities or surprise caused by incongruity.
Meeting abstract presented at VSS 2013