Surprise During Sports Viewing is a Predictor of Behavioural, Physiological and Neuronal Changes

Post by Amanda McFarlan

What's the science?

Event segmentation theory suggests that humans break down the continuous flow of new experiences into discrete segments that can be used as internal models for predicting future events. When these predictions of future events turn out to be wrong, we often experience surprise. Researchers have proposed that surprises may occur at the boundaries between these discrete event segments, which are reflected by behavioural and physiological changes like pupil dilation and increased neuronal activity. Surprise is also thought to be critical for learning and memory in order to update past beliefs with new information. Yet, scientific methods to measure naturalistic surprise are lacking. This week in Neuron, Antony and colleagues investigated how behavioural and physiological changes associated with discrete event segments relate to surprise during naturalistic sports viewing.

How did they do it?

The authors developed and validated a computational model that computes the probability of a given team winning after each change in possession of the ball during a basketball game. Using this model, they computed surprise as ‘the absolute value (either positive or negative) of the change in win probability at each possession boundary’. The authors tracked eye movements and performed functional MRI (fMRI) scans in participants while they watched the last 5 minutes of 9 different basketball games from the National Collegiate Athletic Association college basketball tournament. To study surprise in these participants, the authors created two constructs: belief-consistent surprise (surprise associated with a change in win probability that makes the team with the higher win probability even more likely to win) and belief-inconsistent surprise (surprise associated with a change in win probability that makes the team with the higher win probability less likely to win). Then, they tested the correlation between the proportion of perceived possession boundaries recalled by the participants with belief-consistent and belief-inconsistent surprise at those boundaries.

Next, the authors used hidden Markov models (HMMs; a type of statistical model that captures how the probability of a state depends on a previous state) to analyze the participants’ blood-oxygen-level-dependent (BOLD) responses from the fMRI scan in an attempt to identify discrete segments of neural activity that occurred while watching the basketball game. They investigated whether transitions between HMM-identified segments correlated with changes in possession of the ball, with belief-consistent and belief-inconsistent surprise, and with pupil dilation. Finally, the authors looked at neural activity in two brain areas associated with reward, the nucleus accumbens, and the ventral tegmental area, to determine whether a participant’s preference for one team over the other had an effect on neural activity in these regions depending on whether outcomes were positive or negative for their preferred team.   

What did they find?

The authors found that the proportion of perceived possession boundaries endorsed by the participants was significantly correlated with belief-inconsistent surprise, but not belief-consistent surprise, suggesting that segmentation of events is more robust when new and old information is conflicting. Then, the authors determined that the transitions between HMM-identified neuronal activity segments in the visual cortex and to a lesser extent, in the precuneus and medial prefrontal cortex, were significantly correlated with the true changes in possession during the game. These transitions were also significantly correlated with surprise across possessions in the medial prefrontal cortex and with belief-inconsistent surprise in all three areas. Together, these findings suggest that the visual cortex and precuneus are activated with small to moderate time scale changes like possession turnovers while the medial prefrontal cortex is activated during larger time scale events when there is a greater surprise.

Additionally, the authors showed that changes in pupil dilation were significantly predicted by both belief-consistent and belief-inconsistent surprise. Finally, the authors found that neuronal activity in the nucleus accumbens had a marginally significant correlation with surprise when positive outcomes occurred for the subject’s preferred team, but not for surprise in general. Neuronal activity in the ventral tegmental area was significantly correlated with both overall surprise (regardless of whether the participant had a preferred team) and surprise indicating positive outcomes for the preferred team.

What’s the impact?

This study demonstrates that surprises that contradict a current belief (i.e. which team will win) predicted both behavioural and neuronal activity segmentation as shown by an increased perception of possession boundaries, and changes in neural activity. Additionally, the authors found that surprise correlated with increased pupil dilation and activity in areas of the brain associated with reward, suggesting that humans may have evolved to enjoy unpredictability when it is not critical for survival. Together, these findings provide insight into a novel way to investigate surprise using naturalistic stimuli.

Antony et al. Behavioural, Physiological, and Neural Signatures of Surprise during Naturalistic Sports Viewing. Neuron (2020). Access the original scientific publication here.