Perceiving what we can Conceive: An Attractor Network Model of the Visuo-Semantic Junction

An individual’s understanding of an object appears to permeate into their visual perception of it. Behaviourally, this presents as visual identification and discrimination that is facilitated and actively guided by one’s semantic knowledge. From a neural perspective, semantically dependent modulations of visually evoked electrophysiological potentials are observed. Functional imaging and neuropsychological studies have revealed that these neural modulations are coupled with the activity of the Anterior Temporal Lobe (ATL). These findings allude to the ATL as a source of top-down feedback to earlier regions of the visual hierarchy. The valence and magnitude of this effect appears to be sensitive to the ambiguity and richness of the semantic associations, respectively. To assess the precise nature through which semantic knowledge impacts visual processing, a behavioural experiment was conducted. Participants (N = 113) were trained to associate objects with semantic labels before and after their ability to perceptually discriminate the objects was assessed. A participant’s group denoted whether the semantic associations: (1) overlapped between objects and (2) were rich or impoverished with respect to semantic content. Participants did not differ significantly between groups in their sensitivity or response time to discriminate the objects. Despite this, the group means corresponded to previous studies which adopted a similar paradigm. An attractor network model of the visuo-semantic junction was developed to formalise the interaction of visual and semantic processing. Using the behavioural data as a reference, the most consequential free parameters of the model were selected and tuned in a series of grid searches. The performance of the model displayed correspondence to previous literature on visuo-semantic integration, and provided predictions for future experiments.