Abstract
AIM: The maximum likelihood estimation (MLE) model accounts for classical (spatial) ventriloquism (Alais & Burr, 2004, Current Biology). Does temporal ventriloquism occur, and does the MLE model explain it? METHODS: Temporal localisation was first measured unimodally (2IFC). Each interval was defined by auditory white-noise (1000ms) to provide a temporal reference frame. Auditory trials: a brief tone pulse occurred in each interval (jittered around the midpoint). Observers indicated which was perceived earlier in the 1-sec timeframe. Visual trials: brief luminance blobs were shown in each interval. Observers indicated which occurred earlier in the timeframe. Resulting psychometric functions yielded the subjective temporal midpoint and its variance (precision). Audiovisual trials: first interval contained synchronous audiovisual stimuli at the temporal midpoint; the second contained asynchronous stimuli jittered around the midpoint. Subjects again made temporal order judgements. Subjective midpoint and variance were taken from psychometric functions and compared with MLE predictions calculated from unimodal data. RESULTS: In both modalities, subjective temporal midpoints were typically earlier than actual midpoints. Auditory temporal precision was better than visual. Localisation precision depended on stimulus duration (more precise for brief stimuli). In bimodal conditions, data agreed well with MLE predictions, both for temporal localisation and temporal precision. CONCLUSIONS: Temporal mislocalisations of asynchronous stimuli do occur (temporal ventriloquism). These closely matched MLE predictions, being a weighted average of unimodal signals (each weight the inverse of temporal precision). Also consistent with MLE predictions, bimodal temporal discrimination was more precise than unimodal, indicating statistically optimal integration of information.