Sophie Su

I am an ABD PhD student in Psychology at Washington University in St. Louis, working with Dr. Jeff Zacks.
My research investigates how people build, maintain, and update mental representations of ongoing events during naturalistic perception.

I combine behavioral experiments, eye-tracking, computational modeling, and neuroimaging to study whether mental models update incrementally or globally, and what triggers those updates.

Research Program

How Do Mental Models Update During Continuous Experience?

Humans continuously predict what will happen next—while reading, watching movies, or observing everyday actions.
My research asks:

• What information is represented in mental event models?
• What triggers updates to those representations?
• Are updates local (incremental) or system-wide (global)?
• Can we infer internal updating mechanisms from gaze behavior?

Across projects, I treat eye movements as a real-time behavioral window into dynamic cognitive representations.

Current Projects

🧠 Project 1: Inferring Mental Representations from Gaze Patterns

Core Question
Can we recover the content of people’s mental models by analyzing where and when they look?

Approach
I combine high-frequency eye-tracking with vision–language model embeddings (CLIP) extracted from movie frames to predict gaze distributions during naturalistic viewing.

Key Findings

• Inverting movie frames preserves low-level visual statistics but disrupts semantic structure
• Semantic disruption reduces alignment between CLIP embeddings and action descriptions
• Gaze prediction accuracy drops substantially for inverted scenes

Interpretation
Viewers’ gaze patterns are guided by high-level semantic representations, not just visual salience.

📄 Output:
Vision–Language Model Derived Action Semantics Shape Gaze During Movie Viewing

Ongoing Work
Extending this framework to test how event structure, prediction error, and hierarchy modulate gaze.

🔍 Project 2: Predictive Looking Errors and Event Segmentation

Core Question
How do prediction errors shape how people segment continuous experience into events?

Approach
Using everyday activity videos, I analyze anticipatory eye movements and relate prediction failures to computational model error signals.

Key Findings

• Viewers anticipate actors’ hand movements up to 9 seconds in advance
• Prediction errors align closely with event boundaries
• Gaze-based prediction error mirrors computational model error

Interpretation
Event segmentation is driven by prediction failure, not passive perception.

📄 Publication:
Predictive Looking and Predictive Looking Errors in Everyday Activities
Journal of Experimental Psychology: General
Read the paper

Ongoing Work
Testing whether prediction errors trigger incremental updates or global restructuring of mental models.

🧩 Project 3: Incremental vs. Global Updating of Event Representations

Core Question
When something changes in an event, does the mind update only what changed or everything at once?

Approach
I use:

• Controlled reading and memory paradigms
• Computational representations of situation models
• Eye-tracking and neuroimaging data

to compare incremental updating, global updating, and hierarchical hybrid models.

Current Status

• Dissertation project (data collection and modeling in progress)
• Manuscript in preparation

News & Highlights

2025

• Oct: Paper published in Journal of Experimental Psychology: General
• May: Poster presentation at VSS 2025 (CLIP × gaze project)

2024

• June: Passed subject matter exam on global vs. incremental updating
• May: Talk at VSS 2024 (Predictive Looking project)

2023

• Dec: Departmental research grant awarded
• Aug: Master’s thesis successfully defended

Background (Brief)

• PhD (ABD), Psychology — Washington University in St. Louis
  Advisor: Dr. Jeff Zacks
• Graduate Certificate in Quantitative Data Analysis
• Cognitive, Computational, & Systems Neuroscience Pathway
• B.A. Psychology & Economics — Cornell University

Previously:

• Imaging genetics research on dyslexia in Nicole Landi’s Lab (Haskins)
• Undergraduate research with Khena Swallow (Cornell)
• Neuromatch Academy: Student (2020), TA (2021)

Outside the Lab

I write essays in both Chinese and English and bake cakes that double as small-scale science experiments.
You can find my baking portfolio here.