A cortical semantic space integrating fractions and integers

Using 7T fMRI, this study demonstrates that in educated adults, the brain integrates fractions and integers into a unified cortical semantic space within the anterior inferior parietal cortex, organized by numerical magnitude and domain.

The Gist

The Big Question: How Does Our Brain Count?

Imagine your brain is a massive library. For a long time, scientists knew that this library had a special "Math Section." We knew that when you look at the number 5, a specific part of your brain lights up. We also knew that your brain has a "Mental Number Line"—a mental ruler where small numbers are on the left and big numbers are on the right.

But there was a mystery: Where do fractions live on this ruler?

Integers (whole numbers like 1, 2, 3) are natural; babies and even monkeys understand them. But fractions (like 1/2 or 3/4) are a human invention. They are tricky. In school, kids often struggle to realize that 1/2 is actually bigger than 1/3, even though the number "3" is bigger than "2."

The big question this study asked was: Do educated adults have two separate mental rulers (one for whole numbers, one for fractions), or do they have one single, unified ruler where both types of numbers live together?

The Experiment: A High-Tech Brain Scan

To find the answer, the researchers put 20 smart adults (all with science degrees) into a 7-Tesla MRI machine. Think of this as a super-powerful camera that can see the brain in incredible detail, much sharper than a standard MRI.

The Game:
The participants played a fast-paced game. They saw two numbers appear one after another.

• Example: First they saw 2, then they saw 1/2.
• Task: They had to quickly press a button to say, "Is the second number bigger or smaller than the first?"
• The game mixed whole numbers (0, 1, 2, 3, 4, 5) and fractions (1/4, 1/3, 1/2, 2/3, 3/2, 5/2) randomly.

The Findings: One Big Map, Not Two

The researchers looked at the brain activity while the people played the game. Here is what they discovered, using some fun metaphors:

1. The "Distance Effect" (The Tightrope Walk)

In the game, people were slower and made more mistakes when the two numbers were close together (like comparing 2 and 3, or 1/2 and 2/3) than when they were far apart (like 1 and 5).

• The Metaphor: Imagine walking on a tightrope. It's easy to tell the difference between the start of the rope and the end. But if you are standing on two planks that are right next to each other, it's hard to tell which one is "higher."
• The Result: This happened for both whole numbers and fractions. This suggests the brain treats them the same way: as points on a single line.

2. The "Brain GPS" (The Mental Map)

The researchers used a technique called Multidimensional Scaling (MDS). Imagine you have a list of 12 cities (the numbers). You ask people to draw a map based on how "far apart" the cities feel to them.

• The Result: The brain didn't draw two separate maps (one for integers, one for fractions). It drew one single, continuous line.
• On this line, 0 was at the far left, 1/2 was in the middle, 2 was further right, and 5 was at the far right. The fractions didn't get their own special "island"; they were integrated right into the main highway of numbers.

3. The "Neural Neighborhood" (The Topographic Map)

This is the coolest part. The researchers looked at the Intraparietal Sulcus (IPS), a specific groove in the back of the brain known as the "Math Center."

• The Metaphor: Imagine the surface of this brain area is like a geographic map of a country.
  • The "towns" at the very front of this brain area represent small numbers (like 0 and 1).
  • As you walk backward across the brain surface, the "towns" represent bigger numbers (2, 3, 4, 5).
• The Discovery: They found that the "towns" for fractions were located in the exact same neighborhoods as the "towns" for whole numbers.
  • The brain cells that love the number 1 also love the fraction 1/2.
  • The brain cells that love 3 also love 3/2.
  • They are neighbors! The brain has built a single "Numerotopic Map" where magnitude (size) is the only thing that matters, regardless of whether the number is a whole number or a fraction.

Why Does This Matter?

This study proves that education rewires our brains.

When we are young, fractions feel alien and confusing. But as we learn math, our brain doesn't just add a new "fraction folder" to the library. Instead, it expands the existing shelf. It takes the concept of "size" and stretches it to include these new, abstract symbols.

By the time we are educated adults, our brain has successfully merged these two worlds. We don't think of 1/2 as a weird symbol; our brain instantly converts it into a "size" that fits perfectly on the same mental ruler as the number 1, 2, or 3.

The Takeaway

Your brain is an amazing adapter. It takes the intuitive concept of "how many" (which we are born with) and seamlessly blends it with the complex, cultural invention of "fractions" (which we learn in school). The result is a single, smooth, continuous mental number line where 1/2 and 2 are just neighbors on the same street.

Technical Summary

1. Problem Statement

The study addresses a fundamental question in cognitive neuroscience: how the human brain represents the meaning of abstract symbolic concepts, specifically within the domain of mathematics. While it is well-established that the brain represents non-symbolic numerosities (e.g., sets of dots) and small integers along a "mental number line" in the intraparietal sulcus (IPS), the neural mechanisms underlying symbolic fractions remain poorly understood.

Key challenges include:

• Conceptual Complexity: Fractions are culturally acquired, counterintuitive concepts (e.g., $1/5 < 1/3$ despite $5 > 3$) that do not follow simple successor relations like integers.
• Integration: It is unclear whether educated adults integrate fractions and integers into a single, unified semantic space (a shared magnitude map) or process them via distinct, domain-specific neural codes.
• Resolution: Previous fMRI studies often lacked the spatial resolution to detect fine-grained topographic maps (numerotopic maps) for symbolic numbers, particularly for fractions.

2. Methodology

Participants:

• 20 healthy, right-handed, native French-speaking adults (mean age 26.4) with science bachelor's degrees.
• Excluded: 2 participants (one due to a temporal lobe cyst, one due to a panic attack) and 1 due to equipment failure.

Experimental Design:

• Task: A serial continuous magnitude comparison task. Participants viewed a sequence of 12 numbers (6 integers: 0–5; 6 fractions: 1/4, 1/3, 1/2, 2/3, 3/2, 5/2) and judged whether the current number was larger or smaller than the previous one.
• Stimuli: Presented for 1s with a 4s inter-stimulus interval. All transitions occurred equally often to avoid predictability.
• Math Localizer: A separate run using a sentence verification task (Math vs. Non-math sentences) to identify individual math-responsive regions without using Arabic numerals.

Imaging Protocol:

• Scanner: 7 Tesla (7T) MRI (Siemens) with a 32-channel head coil.
• Resolution: High-resolution functional data (1.2 mm isotropic voxels) using T2*-weighted EPI sequences.
• Preprocessing: Standard pipeline (fMRIPrep) including motion correction, slice-time correction, and normalization to MNI space. Data remained unsmoothed for analysis to preserve fine-grained patterns.

Analytical Approaches:

• Behavioral Analysis: Efficiency scores (Accuracy/RT) and Multidimensional Scaling (MDS) to map the behavioral representational geometry.
• Univariate fMRI: GLM to identify activation differences between domains (Fractions > Integers) and distance effects.
• Multivariate Pattern Analysis (MVPA):
  • PCA: To identify orthogonal dimensions of variance in voxel activation patterns.
  • Representational Similarity Analysis (RSA): Constructing 12x12 Representational Dissimilarity Matrices (RDMs) to test if neural distance correlates with numerical magnitude distance across and within domains.
  • Searchlight RSA: To localize regions where neural dissimilarity is predicted by cross-domain magnitude distance.
  • Voxel Tuning (Gaussian Fits): Fitting Gaussian tuning curves to individual voxel responses to identify "preferred" numerical magnitudes and map their topographic organization (numerotopy).

3. Key Contributions

• Unified Semantic Map: Provides the first high-resolution evidence that in educated adults, fractions and integers are integrated into a single, continuous cortical semantic space organized by magnitude.
• Cross-Domain Topography: Demonstrates the existence of a "numerotopic" map in the anterior intraparietal sulcus (aIPS) and inferior parietal lobule (IPL) where voxels tuned to specific magnitudes (e.g., 0.25, 1, 2) are spatially organized, regardless of whether the stimulus is a fraction or an integer.
• Hippocampal Involvement: Identifies the hippocampus and parahippocampal cortex as key regions encoding graded magnitude information for both domains, suggesting a role in the explicit mapping of complex symbolic concepts.
• Methodological Advancement: Utilizes 7T fMRI and a demanding continuous comparison task to overcome the limitations of previous studies that failed to detect symbolic tuning in the parietal cortex.

4. Key Results

Behavioral Findings:

• Distance Effect: Participants were slower and less accurate when comparing numbers with small numerical distances, regardless of whether the pair was (Integer-Integer), (Fraction-Fraction), or (Integer-Fraction).
• Unified Geometry: MDS of behavioral efficiency scores revealed a one-dimensional curvilinear structure (a "bent" number line) where fractions and integers were interleaved according to their true magnitude (e.g., 1/2 sits between 0 and 1).

Neural Findings:

• Activation Patterns: Univariate analysis showed that fractions elicited stronger activation than integers in the bilateral IPS, precentral gyri, and frontal regions, likely due to higher visual/semantic complexity.
• PCA Results: Two primary components emerged:
  1. PC1: Separated fractions from integers (domain distinction).
  2. PC2: Organized all numbers by magnitude (0 to 5), mixing fractions and integers.
• RSA & Searchlight: Neural activation patterns in the math network (IPS, pITG, MFG, Hippocampus) showed that neural similarity was driven by numerical distance across domains. For example, the neural pattern for $1/4$ was more similar to 0 than to 4, and more similar to $1/3$ than to $5/2$.
• Voxel Tuning & Numerotopy:
  • Individual voxels in the aIPS/IPL showed reliable Gaussian tuning curves for specific magnitudes.
  • Topographic Map: A clear spatial gradient was observed: voxels preferring small numbers were located anteriorly, while those preferring larger numbers were located posteriorly and dorsally.
  • Cross-Domain Consistency: Voxels tuned to a specific integer (e.g., 1) also showed peak activation for fractions of similar magnitude (e.g., 1/2, 2/3), confirming a shared magnitude code.
  • Correlations: Strong positive correlations were found between preferred integer and preferred fraction values within the aIPS/IPL ($r=0.24, p<0.001$), but not in other regions like the MFG or pITG.

5. Significance

• Cognitive Architecture: The findings support the hypothesis that education transforms the brain's representation of numbers. While children may struggle to integrate fractions, educated adults map them onto a pre-existing "mental number line" in the parietal cortex, effectively unifying intuitive (integers) and cultural (fractions) concepts.
• Neural Mechanism: The study confirms that the anterior intraparietal sulcus (aIPS) acts as a hub for a domain-general magnitude representation, capable of encoding abstract symbolic quantities topographically.
• Hippocampal Role: The significant involvement of the hippocampus suggests that for complex, less intuitive concepts like fractions, the brain may rely on explicit memory systems to establish the magnitude relationships before they become fully automated in the cortex.
• Future Directions: This work establishes a framework for investigating how other abstract mathematical concepts (e.g., negative numbers, complex numbers) are integrated into cortical semantic maps and highlights the utility of 7T fMRI for mapping mesoscopic cognitive topographies.