How Action Performance Influences Segmentation and Memory
Research Question
Does mimicking actions while watching affect how people segment events and remember what they saw?
Overview
This project investigates whether action performance—specifically, mimicry during observation—influences event segmentation and subsequent memory. Using a two-study design with everyday activity videos, I examine:
- How action memory relates to event segmentation
- Whether mimicking actions while watching changes segmentation patterns and memory encoding
Theoretical Background
Embodied Cognition and Event Perception
Traditional theories of event perception emphasize visual and conceptual processing. However, embodied cognition theories suggest that motor simulation plays a key role in understanding actions. This project tests whether:
- Motor engagement during viewing affects event boundary detection
- Mimicry changes the granularity of segmentation (coarse vs. fine boundaries)
- Action execution enhances memory for event structure
Event Segmentation and Memory
Event segmentation theory proposes that:
- Events are encoded as discrete chunks in memory
- Event boundaries serve as retrieval cues
- Better segmentation → better memory
If mimicry affects segmentation, it should also affect memory.
Study Design
This project includes two complementary studies:
Study 1: Action Memory and Segmentation
Goal: Establish baseline relationship between action memory and segmentation ability
Design:
- Participants watch 4 everyday activity videos:
- Making breakfast
- Gardening
- Preparing for a party
- Building with Legos
- Segmentation task: Press button at event boundaries (coarse and fine)
- Memory test: Free recall and recognition for actions
Measures:
- Segmentation agreement with normative boundaries
- Memory accuracy (% actions recalled)
- Correlation between segmentation and memory
Study 2: Mimicry While Watching
Goal: Test whether mimicry causally affects segmentation and memory
Design:
- Participants watch the same 4 videos under two conditions:
- Passive viewing: Watch normally
- Mimicry: Mirror actor’s hand movements while watching
- Segmentation task after each video
- Memory test for actions
Predictions:
- Mimicry increases fine-grained segmentation (more boundaries detected)
- Mimicry enhances action memory (better recall for performed actions)
- Effect stronger for goal-directed actions vs. transitional movements
Video Stimuli
Four everyday activities selected for ecological validity:
1. Breakfast Activity
- Making scrambled eggs, toast, and coffee
- Clear action sequences (crack eggs → stir → pour → cook)
- Duration: ~5 minutes
2. Gardening Activity
- Planting seeds, watering, arranging pots
- Object manipulation and spatial arrangement
- Duration: ~4 minutes
3. Party Preparation
- Setting table, arranging decorations, preparing snacks
- Social context, multiple sub-goals
- Duration: ~6 minutes
4. Lego Building
- Constructing a model from instructions
- Hierarchical structure (sub-assemblies → final model)
- Duration: ~5 minutes
All videos filmed from observer perspective to facilitate mimicry.
Methodology
Performance Extraction (Python)
Action performance was coded and extracted using custom Python scripts:
# Pseudocode for performance extraction
def extract_actions(video_path, segmentation_data):
"""
Extract action units from video based on segmentation
Returns:
- action_list: Sequence of action labels
- timestamps: Start/end times for each action
- features: Visual features per action
"""
# 1. Load video frames
# 2. Align with segmentation boundaries
# 3. Extract action units
# 4. Code action types (reach, grasp, move, etc.)
Statistical Analysis (R)
Mixed-effects models testing:
- Effect of mimicry on number of boundaries detected
- Effect of mimicry on memory accuracy
- Individual differences in mimicry benefit
- Correlation between segmentation changes and memory improvements
Preliminary Findings
Study 1 (Action Memory & Segmentation):
- Positive correlation between segmentation agreement and memory accuracy
- Participants who segment more consistently also remember more actions
- Effect strongest for goal-directed actions (vs. transitional movements)
Study 2 (Mimicry):
- Data collection in progress
- Preliminary data suggests mimicry increases fine-grained segmentation
- Mixed results for memory enhancement (may depend on individual differences)
Analysis Pipeline
The act_seg_mem repository includes:
Scripts:
performance_extraction.py: Extract action units from video and segmentation datasegmentation_analysis.R: Analyze boundary detection and agreementmemory_analysis.R: Test memory accuracy and correlationsmimicry_comparison.R: Compare passive viewing vs. mimicry conditions
Data Structure:
data/
segmentation/
coarse_boundaries.csv
fine_boundaries.csv
memory/
recall_data.csv
recognition_data.csv
mimicry/
passive_condition.csv
mimicry_condition.csv
Challenges and Solutions
Challenge 1: Coding Action Units
Problem: Difficult to objectively segment continuous action into discrete units Solution: Multiple coders + hierarchical coding scheme (low-level movements → high-level goals)
Challenge 2: Individual Differences in Mimicry
Problem: Participants vary widely in mimicry fidelity Solution: Record mimicry with video + code fidelity as covariate
Challenge 3: Memory Test Design
Problem: Free recall may be influenced by verbal ability, not just memory Solution: Add recognition test + action ordering test
Implications
If mimicry enhances segmentation and memory:
- Embodied cognition: Motor simulation is not epiphenomenal but functional
- Learning applications: Active imitation may improve skill acquisition
- Clinical applications: Mimicry training for individuals with action understanding deficits (e.g., apraxia)
Future Directions
- fMRI version: Neural correlates of mimicry-enhanced segmentation
- Motor cortex stimulation: Causal role of motor cortex in event perception
- Developmental studies: When does mimicry start influencing event perception?
- Individual differences: Relate mimicry benefit to motor imagery ability
Related Projects
- Predictive Looking: How prediction errors drive segmentation
- Gaze Entropy & Event Boundaries: Proactive cognitive control during viewing
- Incremental vs. Global Updating: How event models are updated
Keywords: Action Perception, Memory, Event Segmentation, Embodied Cognition, Mimicry, Motor Simulation, Everyday Activities
Code & Resources
act_seg_mem
Action segmentation and memory analysis codes for two-study paradigm
