Research

My research focuses on how children and adults use prediction error to guide their reasoning and learning. Prediction sits at the center of how we process information in the world: we form expectations based on our beliefs about the world, notice when reality departs from those expectations, and use prediction error to revise what we believe and guide what we learn next. My work asks how this cycle unfolds: how predictions are formed, how prediction errors are detected and interpreted, and how theses errors change or open people’s minds.

I take an interdisciplinary approach to studying the role of prediction error in human inference, using both behavioral and computational methods, studying both children and adults, and examining both typical and atypical development, such as in autism.

Prediction error guides belief revision

Absence as evidence for hypothesis updating in dynamic prediction tasks

Keywords: prediction, hypothesis updating, absence as evidence

For many people, prediction feels automatic: we know where a ball will roll, or what a friend will do when she picks up her keys. For autistic people, however, prediction may work differently, especially in social situations. Currently, I’m exploring the foundations of prediction in autistic and typically developing children and adults regarding physical objects’ movements. Specifically, we are interested in how people use the absence, rather than the presence, of direct visual information about a target object at an expected time and location. For instance, if you see a ball roll under a table, you might expect it to reappear on the left side. But after waiting a while and not seeing it reappear there, the mismatch between expectation and reality prompts you to consider alternative hypotheses, such as the ball bouncing off a table leg and rolling in a different direction.

Detecting suspicious patterns and finding reasons: children’s representations of coincidence

Keywords: coincidence, causal reasoning, explanation, belief revision

Ever shared a birthday with a colleague or bumped into an old friend in the most unexpected place? Coincidences like these intrigue us and make us wonder, “Why did this happen?” In this project, we seek to understand how adults and children represent coincidences, with a particular focus on the role of available explanations and the number of co-occurrences in shaping these judgments. Our results suggest that representations of coincidence are present early in life, though they continue to develop through the early school-age years.

Check out our paper here.

Prediction error guides selective learning

Learning after observing unexpected behavior from people

Keywords: surprise-induced learning, social learning, violation of expectations

Imagine being at a gathering and someone’s unusual clothing catches your attention. Do you want to learn more about this person, or would you rather avoid them?

As children navigate the social world, they constantly encounter situations like this. When they observe a violation of social expectations, does it enhance their learning, or hinder it?

In a series of experiments, we explore the learning consequences of witnessing violations of expectation in the social domain. Infants (15–19 months old) watched a person commit a violation of psychological expectations (e.g., inconsistent goals, inefficient action, inappropriate emotion), and were then taught either about the person who committed the violation or about the object involved in it. Our data so far suggest that infants learn more about the object involved in the violation, while learning about the agent shows a more complex pattern.

Check out our paper here.

Optimal surprise for learning

Keywords: violation of expectation, surprise-induced learning, explanation

Imagine a ball passing through a solid box. It’s surprising, but you can quickly rationalize it with a reasonable explanation (say, a hidden trapdoor). Now picture that same ball floating in midair after emerging from the box — more puzzling, isn’t it? And if the ball defies even more laws of physics, would you eventually just dismiss it as magic?

Surprise, and its cognitive consequences, is likely graded in nature and closely tied to our efforts to make sense of the unexpected. In this project, we manipulated the number of violations participants observed and examined how surprise, learning, and explanation-seeking behavior changed as a result. We tested adults’ surprise and learning after they observed violations of core physical principles. Our results show that adults’ surprise responses are graded, that surprise is linked to how difficult it is to generate a satisfying explanation, and — interestingly — that a moderate amount of surprise promotes the best learning, suggesting a “Goldilocks” rule of learning.

Check out our CogSci paper here.

Prediction in social reasoning

Predicting others’ competence and performance

Keywords: competence, performance, constraints, reasoning

People are one of the most important information sources children have. Children want to learn from the more competent and knowledgeable person, yet what people do isn’t always reflective of what they’re capable of doing. In the face of this competence-performance gap, children are challenged to predict a person’s actual performance by considering both their competence and other factors that might influence their performance in the moment. Sometimes children’s predictions about someone’s performance turn out to be wrong, prompting them to revise their initial belief about how competent that person is.

In this project, we explore children’s inference of others’ competence and performance, particularly how external constraints affect it. Our general approach is to ask children to predict how people will perform under different circumstances, such as having blocked vision or being distracted by a phone call. We find that even 4-year-olds can make correct inferences in a forced-choice task. In addition, even 20-month-olds already expect that being distracted by a phone call will impair someone’s performance.

Check out our paper here, with another currently in preparation.

Other research interests

I’m also interested more broadly in improving the tools we use to study human cognition, especially in infants, who are tricky to study yet hold the key to so many mysteries of human intelligence. This includes the following methods-focused projects.

Using preferential looking to detect violation-of-expectation effects

Keywords: violation of expectation, preferential looking, methods

The violation-of-expectation (VoE) paradigm has been widely used to study infants’ conceptual understanding. In most VoE studies, expected and unexpected events are presented sequentially. While this allows infants to focus on one video at a time, looking time can be contaminated by order effects.

We are currently testing an alternative approach using a preferential-looking paradigm, in which two stimuli are presented simultaneously on the left and right sides of a screen. We want to see whether infants show a preference between expected and unexpected videos, and what factors shape that preference across trials. Stay tuned!

Individual differences in habituation and dishabituation

Keywords: individual differences, habituation, dishabituation, large-scale datasets

Habituation and dishabituation allow learners to filter out repetitive information and orient toward novelty. Past studies have linked variability in these processes to differences in later cognitive outcomes. Here, we used large-scale datasets from prior studies to examine how individual differences in habituation predict the magnitude of dishabituation in infants, preschoolers, and adults. We found that faster habituation and higher volatility predicted stronger dishabituation. We also found that different measures of dishabituation sometimes produce divergent patterns, suggesting that measurement choices can meaningfully shape observed effects and should be considered carefully in developmental research.

Check out our CogSci paper here.

What does surprise reflect? Conceptual vs. perceptual factors in dishabituation

Keywords: surprise, habituation, dishabituation, looking time

When people are surprised and look longer at what they’re viewing, what does that surprise actually reflect? Low-level perceptual features, or something beyond mere perception? Here, we conducted an exploratory analysis of data from a habituation-dishabituation study (e.g., bird → garlic). We compared looking-time changes for the same pair of stimuli presented in opposite change directions (bird → garlic vs. garlic → bird). If people were relying purely on perceptual features, the same pair of objects should elicit the same degree of surprise regardless of direction, since the perceptual distance between them is identical either way. Yet we found that adults did not treat the two directions as simple reciprocal transformations — for instance, they looked longer at a magical appearance than at a disappearance. This asymmetry in looking time suggests that people were weighing non-perceptual factors even in a simple, image-based habituation-dishabituation study.

Check out our CogSci paper here.