Abstract
Discrepancies between signals from the different senses allow sensory systems to detect and correct calibration errors and thus restore coherent perception of the world. However, given two conflicting cues, the brain must infer which sensory modality requires recalibration. Two types of models have been proposed: (a) Reliability-based: Each modality is recalibrated according to its relative reliability; (b) Fixed-ratio: The degree of recalibration of each modality is fixed. We tested audiovisual spatial recalibration while varying visual stimulus reliability. Visual stimuli were clusters of ten Gaussian blobs; reliability was adjusted by varying the SD of blob locations. The auditory stimulus was a broadband noise burst. We required auditory localization thresholds to fall within the range of tested visual thresholds and ensured this was so by measuring unimodal localization thresholds using a two-interval forced-choice (2IFC) procedure. Next, we measured auditory localization biases relative to visual standards (2IFC). The main recalibration experiment consisted of three phases. (1) Baseline: subjects localized single-cue auditory and visual stimuli. (2) Recalibration: subjects were presented with bimodal stimuli with a perceptually, rather than physically, fixed spatial discrepancy — based on the previously measured auditory localization biases — between the two modalities. Subjects localized one modality, cued after stimulus presentation, using a visual cursor. (3) Post-recalibration: unimodal localization performance was remeasured. In accordance with a fixed-ratio model, most subjects showed no significant visual recalibration even when visual cues were very unreliable. Surprisingly, many subjects showed increased auditory recalibration with decreasing visual reliability, which cannot be explained by either the reliability-based or the fixed-ratio models. Other subjects showed no change or decreasing auditory recalibration. However, a causal-inference model of cross-modal recalibration, in which recalibration requires an inference of the two cues deriving from a common source, captures the diverse influences of cue reliability on recalibration.