Abstract
Perceptual estimation of a stimulus is often attracted toward the previous stimulus (Fischer & Whitney, 2014). The pattern of this attraction largely follows derivative-of-Gaussian (DoG) shape, exerting strong bias when consecutive stimuli are similar. Contrarily, repulsive sequential bias was also reported when different features were asked (Taubert et al., 2016) or time delay between stimulus and response was zero (Bliss et al., 2017). It still remains unclear how these two opposite effects interact. Existing studies using linear regression (DeCarlo & Cross, 1990; Kashiwakura & Motoyoshi, VSS 2018) to disentangle the effects of stimulus and response could not capture the DoG shape of the biases. Here, we empirically demonstrate that observers’ perceptual estimates in sequential tasks result from additive effects of attractive and repulsive sequential biases. Subjects reported the direction of a random-dot motion stimulus whose direction randomly varied from the direction of previous trial following uniform distribution (range: ±20, 40, 80, or 180°). As expected, subjects’ responses were biased toward the motion direction of previous trial. Importantly, when the biases were represented in a 2D surface as a function of both previous stimulus and response directions, it became apparent that the bias on current trial was systematically repelled away from the stimulus direction of previous trial and strongly attracted toward the response direction of previous trial. Moreover, we found that a summation of two DoG curves representing stimulus repulsion and response attraction fits the data closely, which does not systematically differ from the fitting by 2D Gaussian process regression. Magnitudes of extracted attraction and repulsion biases were considerably larger than magnitudes of observed biases in earlier reports. Our results provide direct evidence that both repulsion from previous stimulus and attraction to previous response concurrently occur even in a general sequence of trials, and effects of two biases on current response are largely additive.
Acknowledgement: UNIST Grant 2.180426.01