NUMERACY PODCAST (good but i need to work on it)

Introduction

Numeracy: Scope and Focus

Numeracy is an umbrella term covering a range of mathematical abilities. These abilities are related, but not identical, and they place different demands on cognition and the brain.

Core components include:

• Number

• Arithmetic

• Geometry

• Fractions

• Word problems

• Algebra

Some of these skills are foundational (number, basic arithmetic), while others are more complex and build on earlier abilities (fractions, algebra, word problems).

The focus here is on foundational numerical and arithmetic skills because:

• They form the basis for later mathematical learning.

• They are the most extensively studied.

• They are easier to isolate and study at the brain level.

• Difficulties in mathematics often originate in these early skills.

Two Main Areas of Focus

The material is organised around two closely related topics:

1. Number processing

2. The development of arithmetic

Number processing refers to how numbers are represented, understood, and manipulated. Arithmetic concerns how these number representations are used to calculate an answer.

An Educational Neuroscience Framework

A key framework is used to understand how mathematical skills develop. Although it originated in research on developmental disorders, it is equally useful for understanding typical development.

The framework describes development across multiple interacting levels.

Behavioural Level

This is the most visible level.

• Mathematical abilities are expressed as observable behaviour: counting, calculating, solving problems.

• Differences at this level are what we measure in classrooms and assessments.

• Typical development and difficulties (e.g. dyscalculia) are first identified here.

Cognitive Level

This level explains why behaviour looks the way it does.

Cognitive factors involved in arithmetic learning can be divided into two types:

Domain-specific factors

These are specialised for mathematics:

• Numerical magnitude processing (understanding quantity)

• Order processing

• Mapping numbers onto symbols

• Transcoding (e.g. spoken numbers ↔ written numbers)

Domain-general factors

These support learning across many domains:

• Working memory

• Executive functions (planning, inhibition, flexibility)

• Attention

• Phonological processing

• Metacognition

• Pattern recognition

Both types are essential. Mathematical learning does not rely on a single “math module”.

Biological Level

This level concerns the brain itself.

• Brain structure (which regions are involved)

• Brain function (how these regions are activated and interact)

• Developmental changes over time

Brain data cannot be interpreted in isolation. Cognitive frameworks are needed to understand what changes in brain activity or structure actually mean.

Environment as a Shaping Force

The environment influences all levels of the framework.

Relevant environmental factors include:

• School instruction

• Teaching methods

• Remediation or special education

• Home learning context

• Cultural tools (number symbols, counting systems)

• Strategies encouraged during learning

These factors shape:

• Cognitive strategies

• Symbol use

• Brain development and functional organisation

Brain development related to number and arithmetic is therefore experience-dependent, not fixed.

Why Focus on Foundational Skills?

There are three main reasons:

1. Foundational numerical skills underpin all later mathematics.

2. They are more tractable for brain imaging because they involve fewer overlapping cognitive processes.

3. Mathematical difficulties typically emerge first at this level.

Understanding early number processing and arithmetic provides the clearest window into both typical and atypical development.

Number processing

Are we “hardwired” for number?

• Number symbols (Arabic digits, Roman numerals) are cultural inventions, historically recent.

• Evidence of very old “proto-number” behaviour exists (e.g., tally marks on ancient artefacts), suggesting humans tracked quantity long before formal notation.

• Key idea: we’re likely not evolved for digits, but we are evolved for perceptual/cognitive systems that can be re-used for number.

2) Experience-dependent plasticity → neural recycling

• Neural recycling: brain circuits that originally evolved for other functions (especially vision) get repurposed during development when children learn number symbols.

• Through schooling and repeated exposure, these networks become more specialised for:

  • recognising number symbols

  • extracting magnitude (how much)

• This makes number processing strongly experience-dependent: instruction, strategies, and symbol exposure shape the network you end up with.

Numerical magnitude processing (non-symbolic foundations)

3) Magnitude exists before symbols (and before language)

Evidence across:

• Animals (e.g., monkeys)

• Infants/toddlers

• Cultures with limited number words (e.g., only words up to ~4)

What these groups can do:

• Compare sets and choose “which is more”

• Do simple approximate operations (e.g., rough addition/comparison)

Core properties of this system:

• Nonverbal: no counting words needed

• Non-symbolic: uses sets (dots/objects), not digits

• Approximate: good for big differences, poor for close ones

  • Example: easier to discriminate 8 vs 16 than 8 vs 10

• This is often described as an approximate magnitude system.

From non-symbolic to symbolic: the mapping hypothesis (classic view)

4) Mapping hypothesis (dominant/older view)

• Children first represent quantities non-symbolically.

• Then they learn symbols:

  • counting words (“one, two, three…”)

  • Arabic numerals (1, 2, 3…)

• Symbols gain meaning by being mapped onto those pre-existing quantity representations.

But… symbolic magnitude seems to matter more for maths

5) Symbolic vs non-symbolic tasks predict maths differently

Typical tasks:

• Non-symbolic comparison: which dot set is larger?

• Symbolic comparison: which digit is larger?

Findings described:

• Both relate to overall maths skill, but symbolic processing correlates more strongly with mathematical achievement.

• Longitudinal data: across primary school timepoints, symbolic measures consistently predict later maths performance.

Clinical link:

• In groups with maths difficulties (e.g., dyscalculia; some genetic disorders), a common pattern is poorer performance on symbolic tasks.

Neural basis: numerical magnitude processing in the brain

6) The key region: intraparietal sulcus (IPS)

• Numerical magnitude tasks reliably activate the IPS (both hemispheres).

• Bilateral activation is common.

Left–right differences (tendency, not absolute):

• Left IPS: relatively more involved for symbolic/exact-like processing

• Right IPS: relatively more involved for non-symbolic/approximate-like processing

7) The distance effect (behaviour ↔ brain)

• Behavioural pattern: the closer two numbers are, the harder the decision.

  • 7 vs 8 (distance 1) = harder than 1 vs 8 (distance 7)

• IPS shows a matching neural signature:

  • more IPS activity for small distances (harder comparisons)

  • less IPS activity for large distances (easier comparisons)

• Interpretation: close magnitudes have more overlapping representations, requiring more neural “disentangling.”

8) Beyond IPS: broader network

Meta-analytic work suggests magnitude processing recruits more than IPS:

• Frontal / prefrontal cortex involvement likely reflects domain-general demands

  • attention

  • working memory

  • control/monitoring during decisions

Developmental trajectory

9) What changes with age?

Constraints: fewer child fMRI studies (hard to scan kids still).

What’s known:

• IPS involvement emerges early (seen already around 4–5 years in number comparison tasks).

• With development, there’s a fronto-to-parietal shift:

  • younger children: more PFC activity (needs more support/effort)

  • older children/adults: more IPS specialisation (more efficient, specialised)

• Increasing left-lateralisation can appear with age, consistent with increasing symbolic competence.

This pattern fits an interactive specialisation style of development:

• not “one area matures alone”

• instead, networks reorganise with experience, becoming more specialised.

Challenge to the “simple mapping” story

10) Simple mapping hypothesis has been challenged

Newer behavioural + neuroimaging results suggest symbolic and non-symbolic representations are not always tightly linked.

Key fMRI decoding logic:

• If symbols and dots share a common magnitude code, then a model trained to tell 2 vs 4 from digits should generalise to dots (and vice versa).

• In the described study:

  • decoding within-format works (digits vs digits; dots vs dots)

  • cross-format generalisation fails → suggests partially separate codes

11) Symbolic estrangement hypothesis

Observed pattern:

• People with higher arithmetic skill show less cross-format generalisation.

• People with lower skill / dyscalculia show more generalisation.

Interpretation:

• With expertise, symbol meanings become defined increasingly by relations among symbols, not by a simple perceptual “set of objects” meaning.

  • Example: “3” means different things depending on context:

    • fraction (3/4), decimal (0.3), exponent (x³), algebraic roles, etc.

• So advanced symbolic knowledge can “pull away” from raw quantity.

Symbolic number processing is not just “quantity”

12) Many components must be integrated

Symbolic number understanding includes:

• Quantity (how much)

• Order (before/after; ordinal structure)

• Place value (42 ≠ 24; digit meaning depends on position)

• Syntax / composition rules (how multi-digit numbers are built)

• Language effects: some number word systems are less transparent (e.g., German “four-and-twenty”), adding learning complexity

Important notes:

• These components show large individual differences already in kindergarten.

• Neural bases of these multiple components in children are still not well mapped (open research area).

Summary

• We’re not evolved for digits; we recycle perceptual networks for symbols. Non-symbolic magnitude is early, nonverbal, and approximate; symbolic magnitude predicts maths best. The IPS is central (distance effect), with a developing fronto→parietal specialisation. Simple mapping from dots→digits is too neat; symbols can become estranged from raw quantity as expertise grows, and symbolic processing includes quantity + order + place value + syntax.

Arithmetic development: strategy change + brain network

1) Two ways to solve arithmetic

When you do arithmetic, you typically use one of two broad approaches:

• Fact retrieval: the answer is pulled straight from long-term memory
  Example: 3×3 = 9 feels instant.

• Procedures: you compute via steps (counting, decomposing, borrowing, etc.)
  Example: 1965 − 28 needs working steps.

Both strategies stay available across life; we just shift which one we rely on depending on skill and problem difficulty.

Strategy development over childhood (Siegler-style change)

2) Development is a shift in strategy use

Typical progression:

1. Counting strategies

   • early: count-all (slow, error-prone)

   • later: more efficient counting (e.g., count-on)

2. Basic number combinations (“facts”)

   • repeated exposure + practice → stored answers (e.g., 6+3, 7+7)

3. Derived-fact strategies (decomposition)

   • break problems into easier chunks using known facts
     Example: 8+7 = (8+2)+5 = 10+5

3) Strategy distribution changes

• Early development: more procedures

• With learning: increased fact retrieval

• For harder/larger problems: procedures come back (even in adults)

Dyscalculia and the “strategy bottleneck”

• A major difficulty in dyscalculia is impaired strategy development:

  • weaker/less efficient procedures and counting

  • slower or less fluent fact acquisition

  • difficulties retrieving facts even once learned

• This creates a knock-on effect: if procedures are slow/fragile, it’s harder to build a stable library of facts.

Cross-cultural effects (often ignored, but it matters)

Arithmetic is learned through education, so instruction shapes strategy profiles.

Examples of instructional influence:

• In some contexts (e.g., Belgium example), finger counting is discouraged, pushing children earlier toward retrieval/abstract strategies.

• Other contexts encourage embodied supports (fingers), which can change:

  • how long procedures remain dominant

  • how quickly retrieval becomes fluent

  • potentially the brain patterns measured in “arithmetic tasks”

Bottom line: differences in schooling can produce different “standard” strategies, so brain data can’t be interpreted without the learning context.

Measuring arithmetic strategies

4) Behavioural measurement (most direct)

• Trial-by-trial strategy reports: after each problem, the child reports how they solved it.

• Lets researchers measure:

  • frequency of retrieval vs procedure

  • efficiency (speed/accuracy) of each strategy

5) The common neuroimaging shortcut (and its problem)

Because reporting strategies inside a scanner is hard, many studies infer strategy from:

• problem size (small = retrieval, large = procedure)

• operation (multiplication = retrieval, subtraction = procedure)

But that’s an assumption and is especially shaky in children, because children may:

• use procedures even on small problems

• retrieve some subtraction facts

• switch strategies across trials

So, strategy needs to be measured—not guessed—if you want clean brain interpretations.

Brain imaging study logic: strategy vs operation

A strong design is to identify each child’s strategy use first, then scan.

Example approach described:

• 4th grade children

• Trial-by-trial strategy assessments

• 2×2 design

  • Strategy: Retrieval vs Procedure

  • Operation: Subtraction vs Multiplication

• Include “non-typical” items to break the usual assumptions:

  • retrieval-like subtraction (8−4)

  • procedural multiplication (4×13)

Key result

• Brain activity was modulated by strategy, not operation.

• In other words: the brain seems to care more about how you solved it than whether it was subtraction or multiplication.

The arithmetic network (not a single “math area”)

6) Arithmetic recruits a distributed network

Arithmetic engages multiple regions across the brain, and their involvement changes with:

• instruction

• age

• expertise

• individual differences in performance

7) Domain-specific vs domain-general within the network

You can map the network onto the broader framework:

Domain-specific (number-focused)

• Intraparietal sulcus (IPS): consistently active in calculation

  • often more active for procedures than retrieval

  • plausible reason: procedures require magnitude/semantic manipulation (decomposing numbers intelligently)

Domain-general (support systems)

• Prefrontal cortex (PFC): attention + working memory demands

  • typically stronger for procedures (multi-step control)

• Hippocampus: memory learning/retrieval involvement

  • fits with retrieving learned associations/facts

• Angular gyrus (AG) / Supramarginal gyrus (SMG) (inferior parietal)

  • linked to semantic retrieval and verbal/learned knowledge (often discussed in relation to fact retrieval)

• Fusiform gyrus (ventral visual stream)

  • symbol recognition (digits like “visual word” style processing)

• Posterior superior parietal/superior frontal regions (spatial attention)

  • sometimes explained via spatial strategies like number-line-like processing

8) Connectivity matters

• Performance correlates with the quality of white matter connections between regions (fronto–parietal links are a frequent example).

• These tracts are not arithmetic-specific (they also relate to skills like reading), which fits the idea that arithmetic depends partly on domain-general infrastructure.

Development of the arithmetic network (what changes in the brain as strategies change)

1) The annoying limitation: not much longitudinal data

• There’s little/no true longitudinal fMRI tracking the same kids across many years.

• So most “development” evidence is based on:

  • age correlations (older kids vs younger kids)

  • cross-sectional comparisons (groups of different ages)

• Useful, but it’s not the cleanest way to infer how individual brains change over time.

A) What cross-sectional development tends to show

2) Network specialisation with age (interactive-specialisation vibe)

A common pattern when comparing younger children → adolescents/adults:

• Decreases in prefrontal cortex (PFC) activity with age
  → interpreted as reduced reliance on effortful control, working memory, and “doing it the hard way”.

• Increases in parietal activity with age
  → interpreted as increasing specialisation/efficiency in number–arithmetic representations (parietal systems doing more of the heavy lifting).

This fits the bigger idea: development isn’t one area maturing alone; it’s the network reorganising as skill improves.

3) Fact retrieval might be “graded” (not one single state)

The slides hint at a staged view of retrieval:

• Early consolidation stage: retrieval still leans on the hippocampus

  • Think: the fact is newly learned, fragile, still being “installed” into cortex.

• Automatised stage: retrieval relies more on angular gyrus (AG) / inferior parietal regions

  • Think: the fact is now a stable, quickly accessible long-term representation.

So retrieval isn’t just “on/off”—it may shift from hippocampal-supported → cortical/AG-supported as facts become automatic.

B) Training studies: a window into development (because we can manipulate learning)

4) Why training studies are useful

• You manipulate the learning process, not just observe it.

• That lets you ask more causal questions like:

  • “If we push items from procedures → retrieval, what changes in the network?”

This is especially handy when long-term developmental data is scarce.

Adult training pattern (the classic result)

When adults learn new arithmetic facts:

• Early learning (untrained items) shows more activity in:

  • fronto–parietal control/magnitude systems (effortful computation)

• After training:

  • fronto–parietal activity decreases

  • angular gyrus activity increases (more “fact-like” retrieval)

That’s the clean “effortful → automatic” signature.

C) Child fMRI training study (the one in your slides)

5) Design (what they did)

• Children around 10 years old (mean ~10.3)

• Two scan sessions:

  • fMRI 1 (before)

  • training across days

  • fMRI 2 (after)

• Training task: multidigit multiplication

• Included:

  • trained problems (TR) vs untrained problems (UN)

Training produced the expected behavioural learning curve:

• accuracy ↑

• reaction time ↓

6) Brain results: procedures → retrieval transition is clear

Main robust effect:

• After training, trained problems show reduced activity in:

  • frontal regions (less working memory / control load)

  • parietal regions including IPS (less magnitude-manipulation / stepwise computation)

Interpretation:

• Training makes the “new” problems start to resemble already-known problems.

• This looks like a shift away from procedure-heavy solving.

So the development of procedural efficiency / reduced effort is pretty consistent and shows up reliably.

7) But “development of retrieval” is messy

The tricky part:

• The neat adult pattern (AG up, clean retrieval signature) doesn’t show up as clearly in children.

Instead:

• Retrieval-related reorganisation is described as complicated / not captured cleanly.

• There’s mention of hippocampal involvement changing with training, but no stable, simple pattern emerges.

Why might it be messy (conceptually consistent with the slides)?

• In children, “retrieval” may still be partly hippocampal-supported (facts not fully automatised).

• The same child might mix strategies even after training (partial retrieval + fallback procedures).

• “Retrieval” may not be one process: it can mean newly learned fact recall vs fully automatised fact access.

The core takeaways to remember

• Developmental evidence is limited; much comes from age comparisons, not true longitudinal tracking.

• With age/skill, arithmetic networks tend to shift toward greater specialisation:

  • PFC down, parietal up

• Fact retrieval may be graded:

  • early retrieval ↔ hippocampus

  • automatised retrieval ↔ angular gyrus

• Training studies show a reliable signature of becoming more efficient:

  • fronto–parietal (incl. IPS) activity decreases as items become learned

• The exact neural “signature” of retrieval maturation in children is less settled than the procedural-efficiency story.

That last point is a classic neuroscience mood: procedures cleanly visible; retrieval is a sneaky shape-shifter.

Challenges and future directions

1) Truly understanding development requires better data

• The biggest limitation in this field is the lack of longitudinal studies.

• Most claims about “development” are based on:

  • age correlations

  • cross-sectional comparisons between groups

• These approaches cannot tell us how the same child’s brain changes over time.

• Future progress depends on tracking within-individual change, not just age differences.

2) Shift from descriptions to mechanisms

Much of the existing work is descriptive:

• “This region is active”

• “This network changes with age”

What’s still missing:

• Clear mechanistic explanations

• Separation of causes ↔ consequences

  • Does brain change drive learning?

  • Or does learning drive brain change?

Understanding mechanism requires designs that manipulate learning, not just observe it.

3) Interventions as scientific tools (not just applications)

• Training and intervention studies should not only aim to “improve maths”

• They are crucial for:

  • testing causal hypotheses

  • isolating which learning experiences change which neural systems

• Many existing “training” studies are experimental simulations, not realistic educational interventions

• There is a need for theory-driven, educationally valid interventions linked to brain measures

The educational environment matters (a lot)

4) Education is not a nuisance variable

• Arithmetic learning happens in structured educational contexts

• Yet educational context is often:

  • poorly characterised

  • ignored in cognitive neuroscience interpretations

Key contextual factors include:

• emphasis on counting vs retrieval

• focus on fluency vs conceptual understanding

• use (or discouragement) of tools like fingers

• instructional pacing and practice intensity

These factors moderate both behaviour and brain activity.

5) Cross-cultural differences are informative, not noise

• Different countries emphasise different instructional priorities

  • counting-based strategies

  • speed and fluency

• These differences change:

  • which cognitive predictors matter most

  • which neural networks are recruited during arithmetic

• Conclusion: there is no single “universal” arithmetic brain

  • the brain adapts to how arithmetic is taught

How does formal schooling change the brain?

6) The school cut-off approach

A powerful quasi-experimental method:

• Compare children of nearly the same age

• One group has started school, the other has not

• Differences can be attributed to schooling, not maturation

This approach allows stronger causal claims about schooling effects on:

• maths

• reading

• brain structure and function

7) Math and reading do not develop spontaneously

Findings from school cut-off studies show:

• Clear gains in:

  • number comparison

  • number naming

  • ordering

  • calculation

  • phonological awareness

  • letter knowledge

  • reading

• These gains align with school exposure, not age alone

Schooling actively constructs these abilities.

8) Possible effects on brain structure (early evidence)

• Preliminary data suggest schooling may influence:

  • cortical thickness in parietal regions

  • language-related frontal regions

• These findings are early and tentative

• Still, they support the idea that formal education leaves a biological trace

Core conclusions to hold onto

• Numerical magnitude processing is reliably linked to intraparietal sulcus activity (with frontal support).

• Symbolic magnitude processing is most critical for mathematical success, but its developmental origins remain unresolved.

• Arithmetic relies on a distributed network, not a single “math area”.

• Brain activity during arithmetic is modulated by strategy use and changes with learning.

• To move forward, research must:

  • take development seriously

  • use longitudinal and training designs

  • explicitly model the educational context

• Educational environments vary, and these variations shape both cognition and brain organisation.

Numeracy Development – Revision Notes

1) Neural networks for number magnitude processing and arithmetic

Number magnitude processing

• Core region: Intraparietal Sulcus (IPS) (bilateral)

  • Represents numerical magnitude

  • Shows the distance effect (closer numbers → more activity)

• Left IPS: more involved in symbolic / exact processing

• Right IPS: more involved in non-symbolic / approximate processing

• Frontal regions (PFC) support attention and working memory, especially in children

Arithmetic network (distributed)

• IPS: magnitude manipulation, especially during procedures

• Prefrontal cortex (PFC): cognitive control, working memory (procedures > retrieval)

• Angular gyrus (inferior parietal): automatised fact retrieval

• Hippocampus: early fact learning and consolidation

• Fusiform gyrus (ventral stream): visual recognition of number symbols

• White-matter tracts (fronto-parietal): connectivity quality correlates with performance

Key principle: arithmetic relies on a network, not a single “math area”.

2) Developmental trajectory of numeracy skills and links to maths competence

Early foundations

• Infancy: non-symbolic, approximate magnitude processing (no language, no symbols)

• Preschool (≈3–5 yrs): learning to map symbols (number words, digits) onto quantities

• IPS activity for number comparison emerges early (~4–5 yrs)

Schooling effects

• Formal schooling drives rapid gains in:

  • symbolic magnitude processing

  • ordering

  • calculation

• Numeracy does not develop spontaneously without instruction

Arithmetic strategy development

• Early: counting and procedural strategies

• Middle childhood: increasing fact learning and derived-fact strategies

• Later: greater reliance on fact retrieval, with procedures used for harder problems

Brain development

• Fronto → parietal shift with age and learning:

  • PFC activity decreases (less effortful control)

  • Parietal specialisation increases

• Fact retrieval may be graded:

  • early retrieval ↔ hippocampus

  • automatised retrieval ↔ angular gyrus

Relation to mathematical competence

• Symbolic magnitude processing is a stronger predictor of maths achievement than non-symbolic processing

• Strategy efficiency (retrieval vs procedure) strongly predicts performance

• Educational context (instruction style, fluency emphasis) moderates both behaviour and brain activity