Abstract
The adult human visual system robustly recruits stimulus position (i.e. visual direction) as a cue for disambiguating the apparent rotation directions of 3D figures, as the result of training during which position is paired with binocular disparity and occlusion (Haijiang, Saunders, Stone & Backus, 2006). How should we measure the system's reliance on the newly recruited cue? Backus & Haijiang (2007) used conflict stimuli to quantify reliance in units of disparity, but one can also measure reliance in units of decision noise using probit analysis (Dosher, Sperling & Wurst, 1986). We hypothesized that reliance on the new cue would generalize completely from binocular to monocular viewing. Each of six trainees viewed 480 training stimuli and 480 test stimuli in alternation. 2/3 of the test stimuli contained small binocular disparities that agreed or conflicted with the new position cue. The remaining test stimuli were monocular. The the new cue was differently effective across trainees, from 0.6 to 5.8 normal equivalent deviations (NEDs), but had similar effects for monocular and binocular stimuli within each trainee. Across trainees, mean difference ± SE was 0.1±0.4 NED, with r = 0.89 (p [[lt]]0.02). The results are consistent with a model in which the visual system chooses one or the other perceptual interpretation according to subjective reliabilities, determined by the cues, that act like Bayes factors. Additivity of effects is predicted by identifying the cues either with log likelihoods or with “equivalent Bernoulli experiments” to estimate a posterior Bernoulli probability. Explaining perceptual bistability differs from explaining rivalry because one can ignore the time course of decision making. Unlike rivalry, binary perceptual decisions can easily be treated within existing theoretical frameworks for cue combination, which makes bistable stimuli highly appropriate for testing quantitative models of cue recruitment and cue combination.
NIH R01-EY013988, NSF 617422.