An implicit representation of stimulus ambiguity in pupil size

Abstract

To guide behavior, perceptual systems must operate on intrinsically ambiguous sensory input. Observers are usually able to acknowledge the uncertainty of their perception, but in some cases, they critically fail to do so. Here, we show that a physiological correlate of ambiguity can be found in pupil dilation even when the observer is not aware of such ambiguity. We used a well-known auditory ambiguous stimulus, known as the tritone paradox, which can induce the perception of an upward or downward pitch shift within the same individual. In two experiments, behavioral responses showed that listeners could not explicitly access the ambiguity in this stimulus, even though their responses varied from trial to trial. However, pupil dilation was larger for the more ambiguous cases. The ambiguity of the stimulus for each listener was indexed by the entropy of behavioral responses, and this entropy was also a significant predictor of pupil size. In particular, entropy explained additional variation in pupil size independent of the explicit judgment of confidence in the specific situation that we investigated, in which the two measures were decoupled. Our data thus suggest that stimulus ambiguity is implicitly represented in the brain even without explicit awareness of this ambiguity.

Keywords: ambiguity; auditory perception; confidence; pupillometry; uncertainty.

Fig. 1.

Behavioral responses to ambiguous frequency shifts. (A) Schematic showing task design and timing on each trial. After a variable baseline period of 3 to 5 s, participants heard two tones (T1 and T2, 250 ms each) separated by 250 ms. At the end of the waiting period (4 s after T1 onset, cued by offset of background noise), participants made a keypress response categorizing the change between T1 and T2. (B) Schematic spectrograms showing stimuli in each condition of Experiment 1. Dashed lines indicate the shortest path between components, with upward and downward paths equally prominent at 6 and ±2 ST. (C) Probability of each behavioral response by condition in Experiment 1. Responses "up" and "down" are equivalently frequent at 6 and ±2 ST, but the response "both" is more frequent at ±2 ST. Error bars show ±1 SEM (n = 20).

Fig. 2.

Pupil dilation in Experiment 1. (A) Pupil dilation throughout the trial, baseline corrected from 100 ms before T1 onset to T2 onset, z-scored for each subject, and averaged across subjects in each condition. For plotting purposes, 2 and 10 ST are combined, and 4 and 8 ST are combined, as these conditions have equivalent interval sizes (note that they were separate conditions in our analyses). Gray regions indicate the presentation of T1 and T2. The dashed vertical line indicates the onset of the response window. The horizontal gray bar indicates the temporal cluster in which the main effect of condition on pupil size was significant. Colored regions show ±1 SEM (n = 20). (B) Pupil dilation (averaged over the identified cluster) for each interval size. (C) Entropy of behavioral responses for each interval size. Error bars show ±1 SEM (n = 20).

Fig. 3.

Pupil dilation correlates with response entropy and interval size for ambiguous frequency shifts. (A) Time course of the entropy-related pupil response in Experiment 1, shown as beta weights from sample-by-sample regressions of response entropy on pupil size. The horizontal bar indicates P < 0.05 in a cluster-corrected permutation test against zero, and the colored region shows ±1 SEM (n = 20). (B) Time course of the interval size-related pupil response, as in A but replacing response entropy with interval size. (Center) Individual coefficients for response entropy (C) and interval size (D) in two separate Bayesian hierarchical models of mean pupil dilation from 2.06 to 4 s in Experiment 1. (Right) Comparison of predictions from the response entropy (E) and interval size (F) models and observed data. Each subject’s individual mean has been subtracted from both model predictions and observed data in order to focus on within-subject differences. Each circle shows one condition for one listener, with the associated response entropy or interval size shown by both the color and the size of the circle.

Fig. 4.

Behavioral responses to ambiguous and unambiguous frequency shifts. (A and B) Schematic spectrograms of stimuli in Experiment 2, with dashed lines indicating shortest paths between components. The DL is the listener’s individually measured frequency DL. Ambiguity emerges at 5.5 ST for Shepard tones but not for Harmonic complexes. (C and D) Probability of responding "up" and average confidence rated on a four-point scale for each condition in Experiment 2. Behavioral variability is greatest for 5.5 ST Shepard tone intervals and smallest for 5.5 ST harmonic complexes. Confidence is lowest at the DL for both tone types. Colored regions show ±1 SEM (n = 25). (E and F) Response entropy and confidence for each tone type and interval size, collapsing across interval sign. Error bars show ±1 SEM (n = 25).

Fig. 5.

Pupil dilation in response to ambiguous and unambiguous frequency shifts. (A and B) Time course of pupil dilation in Experiment 2, baseline-corrected, z-scored, and averaged across subjects in each condition. Gray regions indicate T1 and T2 presentation. The dashed vertical line indicates the onset of the response window. Gray horizontal bars indicate P < 0.05 in a cluster-corrected permutation test, in dark gray (1) for the interaction of interval size by tone type on pupil size, and in light gray (2) for the main effect of tone type on pupil size. Colored regions show ±1 SEM (n = 25). (C and D) Pupil size in each condition averaged within the interaction cluster (1) and the tone type cluster (2). Error bars show ±1 SEM (n = 25).

Fig. 6.

Pupil dilation correlates with response entropy. (A) Time course of the entropy-related pupil response in Experiment 2, shown as beta weights from sample-by-sample regressions of response entropy on pupil size. The horizontal bar indicates P < 0.05 in a cluster-corrected permutation test against zero, and the colored region shows ±1 SEM (n = 25). (B) Comparison across conditions of response entropy and mean pupil dilation in the overall cluster previously identified in Experiment 1 (2.06 to 4 s). Error bars show ±1 SEM. (C) Time course of the confidence-related pupil response, as in A but replacing response entropy with confidence ratings, inverted to reflect the expected direction of the effect. No significant cluster was identified. (D) Comparison across conditions of inverted confidence and pupil dilation, as in B. (E) Time course of the interval size–related pupil response, as in A but replacing response entropy with interval size. No significant cluster was identified. (F) Comparison across conditions of interval size and pupil dilation, as in B.

Fig. 7.

Modeling the effects of response entropy and confidence on pupil dilation. (A) Individual coefficients for each of the four effects in a Bayesian hierarchical model of mean pupil dilation from 2.06 to 4 s in Experiment 2. Error bars show ±1 SEM (n = 25). (B) Comparison of response entropy and confidence in Experiment 2. Each circle shows one condition for one listener, with the associated mean pupil dilation shown by both the color and size of the circle. Response entropy and confidence are negatively correlated (r = −0.37), but examples of both high-entropy/high-confidence and low-entropy/low-confidence combinations are numerous. (C) Comparison of predictions from the Bayesian hierarchical model and observed data. Each subject’s individual mean has been subtracted from both model predictions and observed data, in order to focus on within-subject differences. Each circle shows one condition for one listener, with the associated response entropy shown by both the color and the size of the circle. (D) The same as in C but with color and size of circles showing confidence rather than response entropy.