Abstract
Top-down feedback is an important yet poorly understood aspect of visual processing, due to an ambiguity in the involvement of bottom-up and top-down information in the processing of complex stimuli. For example, an apparent categorical or conceptual difference may be derivable from a feed-forward process based on simple visual statistics, particularly in familiar stimuli. Convolutional neural networks have evidenced a striking ability to successfully perform such tasks in the absence of any top-down feedback. Here, we present the first steps in a series of studies aimed at disentangling top-down and bottom-up effects by taking advantage of several unique aspects of visual word recognition. Words afford excellent control of their low-level characteristics, unambiguous involvement of non-visual information, and a high degree of localization of their cortical processing. To begin, we examined the word superiority effect (WSE) wherein letters are more quickly and accurately identified in words than pseudowords, taken to be indicative of top-down feedback. The current experiment is the first in a series of studies that will investigate whether WSE is modulated by semantics and direct physiological intervention. We hypothesized that frequency should modulate the strength of connections between every high-level system and the ventral temporal locus of letter recognition. As predicted, we found a strong correlation between word frequency and the strength of the WSE in both accuracy and reaction time. Further, the location within the US where a person lives was found to modulate the WSE by defining their familiarity with regionally specific words. Other manipulations, such as the addition of cognitive load, as well as priming of both orthographic and semantic neighbors, yielded modulations. These results demonstrate the potential utility of studying top-down feedback in the context of visual word recognition.