Late cortical dynamics mediate the symmetry-induced numerosity illusion

Using EEG, this study demonstrates that while the human brain encodes physical numerosity directly within 50 ms, the symmetry-induced underestimation of object count arises from later grouping processes occurring around 150 ms that integrate spatial arrangement to shape perceived numerosity.

The Gist

Imagine your brain is a busy newsroom trying to count the number of reporters in a chaotic room. Sometimes, the reporters are scattered randomly, and sometimes they are standing in perfect, mirrored lines.

This study is like a high-speed camera filming the newsroom's internal communications (using EEG electrodes on the scalp) to see exactly when and how the brain figures out the count. The researchers wanted to solve a mystery: Does the brain count the people directly, or does it get tricked by how they are arranged?

Here is the story of what they found, broken down into simple steps:

1. The "Raw Data" Arrival (The First 50-80 Milliseconds)

The Analogy: Think of the brain as a super-fast scanner at a security checkpoint.
When the dots (the reporters) first appear on the screen, the brain's "scanner" zips through the image almost instantly. In less than a blink of an eye (about 50 to 80 milliseconds), the brain has already counted the physical number of dots.

• What this means: The brain doesn't need to wait to see if the dots are in a pretty pattern or a messy pile. It grabs the raw number immediately, ignoring other tricks like how big the dots are or how much space they take up. It's like a cashier scanning a barcode before you even finish putting the item on the belt.

2. The "Grouping" Glitch (Around 150 Milliseconds)

The Analogy: Now imagine the reporters in the newsroom start organizing themselves. If they stand in perfect, mirrored lines (symmetry), they look like a single, unified team rather than a crowd of individuals.
About 150 milliseconds after the image appears, the brain's "editor" steps in. This editor looks at the arrangement and says, "Oh, these are grouped together!"

• The Illusion: Because the dots are grouped by symmetry, the brain decides to count the groups instead of the individuals. It's like looking at a bouquet of flowers and counting the bouquets instead of the petals.
• The Result: The brain starts to underestimate the number. A symmetrical array of 24 dots starts to feel like it has fewer dots (maybe around 22) because the symmetry makes them look like a single, cohesive object.

3. The "Perceived" Number (170-240 Milliseconds)

The Analogy: This is the moment the news headline is written.
By the time the brain finishes its initial scan and its grouping analysis (around 170 to 240 milliseconds), the "official count" is updated. The brain is no longer reporting the raw physical number; it is now reporting the perceived number.

• The Twist: If you asked a person at this exact moment, "How many dots are there?" they would likely say a lower number for the symmetrical pattern than for the random one, even though the physical count is identical. The brain has been "fooled" by the pattern, but only after it did the initial, accurate count.

The Big Takeaway: Two-Step Processing

The study reveals that our brain has a two-step process for counting:

1. Step 1 (The Fast Lane): A lightning-fast, automatic scan that gets the true, physical number right away. It's honest and direct.
2. Step 2 (The Slow Lane): A slightly slower process that looks at the context and patterns. It groups things together based on symmetry. This step overwrites the first count, leading to the illusion that there are fewer items when they are neatly arranged.

Why Does This Matter?

This is like realizing that your brain is a two-stage detective.

• Stage 1 is the rookie cop who counts every suspect immediately.
• Stage 2 is the seasoned detective who looks at the suspects' uniforms and realizes, "Wait, they are all wearing the same uniform; they must be one gang." The seasoned detective then changes the report to say "One gang" instead of "Ten individuals."

The study proves that the "mistake" (the illusion) isn't a failure of the brain's counting ability; it's a feature. The brain is so good at finding patterns and grouping things together that it sometimes sacrifices the exact count to understand the structure of the scene. This helps us navigate the world efficiently, even if it occasionally tricks us with a few extra dots!

Technical Summary

1. Problem Statement

The study addresses two fundamental questions in visual neuroscience regarding numerosity perception (the ability to estimate the number of objects):

1. Direct Encoding vs. Inference: Is numerosity encoded directly by dedicated neural mechanisms, or is it inferred from correlated non-numerical features (e.g., total area, density, item size, convex hull)? While fMRI studies suggest direct encoding, their low temporal resolution cannot distinguish between feedforward encoding and later feedback processes.
2. Integration of Contextual Cues: How and when does the brain integrate physical inputs with spatial organizational cues (specifically symmetry) that bias perception? Previous behavioral work shows that symmetrical dot arrays are perceived as less numerous than random arrays (the symmetry-induced underestimation illusion), but the neural timeline of this integration remains unknown.

2. Methodology

The researchers employed Electroencephalography (EEG) combined with advanced multivariate analysis techniques to track neural dynamics with high temporal resolution.

• Participants: 19 healthy adults (after exclusions).
• Stimuli: Central dot arrays with varying numerosities (10, 12, 22, 24 dots).
  • Conditions: Dots were arranged either symmetrically (relative to vertical/horizontal axes) or randomly.
  • Control: Low-level visual features were rigorously controlled. In half the trials, convex hull was equated; in the other half, density was equated. Item size was manipulated to match total surface area where necessary.
• Task: Participants viewed stimuli for 150ms. On 10% of "catch trials" (red stimuli), they compared the current numerosity to the previous one. These trials were excluded from analysis to ensure attention without confounding the passive viewing data.
• Analysis Techniques:
  1. Univariate Analysis (VEPs): Linear mixed-effects models on Visual Evoked Potentials to identify main effects of numerosity and symmetry over time.
  2. Multiple Regression Representational Similarity Analysis (RSA): A novel application to EEG time-series. Neural Representational Dissimilarity Matrices (RDMs) were regressed against theoretical RDMs for numerosity, item size, convex hull, and spatial arrangement. This isolated the unique variance explained by numerosity independent of other features.
  3. Perceived vs. Physical RSA: A specific RSA model compared neural data against "physical numerosity" vs. "perceived numerosity" (where symmetric arrays were mathematically down-weighted based on behavioral underestimation estimates).
  4. Multivariate Pattern Analysis (MVPA): Linear Discriminant Analysis (LDA) classifiers were used to decode numerosity and spatial arrangement. Temporal Generalization Analysis determined if neural representations were transient (time-specific) or sustained (stable over time).

3. Key Results

A. Sequential Encoding of Physical vs. Perceived Numerosity

• Early Phase (~50–100 ms):
  • Physical Numerosity: Decodable from occipitoparietal electrodes as early as ~50 ms.
  • Independence: RSA confirmed that this early signal was uniquely driven by numerosity, independent of item size, convex hull, or spatial arrangement.
  • Nature: This represents a rapid, direct encoding of "raw" physical numerosity.
• Late Phase (~150–240 ms):
  • Symmetry/Arrangement: The neural signature of spatial arrangement (symmetry vs. random) emerged later, around 150 ms.
  • Perceived Numerosity: In the 170–240 ms window, neural patterns were better predicted by perceived numerosity (biased by symmetry) than by physical numerosity.
  • Interaction: A significant interaction between numerosity and symmetry was observed in this late window, indicating that the symmetry-induced underestimation effect is realized at this stage.

B. Temporal Dynamics (Transient vs. Sustained)

• Numerosity: Early numerosity decoding was transient (highly time-specific, confined to the diagonal in temporal generalization matrices), suggesting a brief feedforward sweep.
• Spatial Arrangement: Decoding of symmetry was sustained and temporally distributed, generalizing across a broader time window, suggesting a more stable representation that allows for integration with other cues.

C. Comparison with Connectedness Illusion

The temporal profile of the symmetry-induced underestimation (emerging ~150–170 ms) closely mirrors the connectedness illusion (where linked items are perceived as fewer), which also emerges later than physical numerosity encoding.

4. Key Contributions

1. Temporal Dissociation: The study provides the first high-temporal-resolution evidence that the brain processes physical numerosity and perceived numerosity in distinct temporal stages. Physical numerosity is encoded directly and rapidly (50 ms), while the perceptual bias caused by symmetry is a later, integrative process (170 ms).
2. Methodological Advancement: Successfully applied Multiple Regression RSA to EEG data to disentangle numerosity from confounding non-numerical features (convex hull, density) in real-time, overcoming limitations of previous fMRI studies.
3. Mechanism of Illusion: Demonstrates that the symmetry-induced underestimation is not a failure of early sensory encoding but a result of late-stage perceptual grouping. The brain initially counts the dots correctly, then revises this count based on grouping cues that bind symmetric elements into unified objects.

5. Significance

• Neural Architecture of Number Sense: The findings support the "direct encoding" hypothesis for early visual processing, confirming that numerosity is a fundamental visual attribute. However, it refines this view by showing that the final percept is constructed through later, attention-dependent grouping mechanisms.
• Role of Grouping: The study validates the theory that numerosity is computed from perceptually segregated units rather than individual elements. Symmetry acts as a grouping cue that reduces the perceived count by merging elements into fewer "objects."
• Feedforward vs. Feedback: The late emergence of the symmetry effect (post-150 ms) suggests it may rely on incremental grouping mechanisms, potentially involving recurrent feedback loops or multiple feedforward sweeps, rather than purely initial feedforward processing. This parallels findings on the connectedness illusion and highlights the dynamic, constructive nature of visual perception.

In conclusion, the paper establishes a clear temporal hierarchy in numerosity perception: an early, automatic, and direct encoding of physical quantity, followed by a later, context-dependent revision where spatial organization (symmetry) modulates the final perceptual estimate.