Direct Answer: Abstraction works when a learner identifies the relationship that remains useful across changing examples and represents that relationship without depending on the original surface details. The learner moves from “this particular case” towards “this kind of structure.” In Mathematics, this may mean recognising proportionality across different contexts. In Science, it may mean seeing the same conservation relationship across different systems. In English, it may mean recognising a claim–evidence relationship across different topics. Abstraction does not mean stripping away every concrete example. Good abstraction usually depends on examples, comparison and prior knowledge. The aim is to remove irrelevant detail while preserving the structure that makes transfer possible.
HOW LEARNING WORKS · ABSTRACTION
The useful idea is the part that can travel.
Abstraction asks what remains true when the names, numbers, colours, stories and layouts change.
The simplest definition
Abstraction is the process of identifying and representing a general structure by separating essential relationships from incidental surface features.
A learner who knows only the original example has memory of a case. A learner who can state what makes that case an instance of a wider structure has begun to abstract.
The abstraction mechanism
EXAMPLE A → EXAMPLE B → ALIGN FEATURES → IDENTIFY COMMON RELATIONSHIP → REMOVE IRRELEVANT DETAIL → STATE INVARIANT → REPRESENT MORE GENERALLY → TEST AGAINST NEW EXAMPLES → REFINE BOUNDARY → USE IN NEW CASE
Abstraction is therefore not a single leap away from concrete experience. It is often a cycle of comparison, compression and testing.
1. The surface can be memorable and misleading
A learner studies three ratio questions about recipes and concludes that ratios are about food. Another learns electrical resistance from one circuit layout and later fails when the same relationship is drawn differently.
The examples were remembered, but the structure remained attached to their surfaces.
Abstraction begins when the learner can say which features were merely part of the story and which features controlled the reasoning.
2. Comparison is one of the strongest routes into abstraction
Two cases make common structure easier to notice because the learner can align them.
Ask: What is different on the surface? What relationship is the same? Which step would still be valid if the context changed? Which feature decides the method?
Research on analogical encoding has shown that comparing cases can support schema abstraction and transfer in some domains. This does not mean comparison always produces abstraction automatically; the comparison has to direct attention towards the relevant relationship.
3. Abstraction needs enough examples—but not endless examples
One example makes it difficult to know what is essential. Too many unstructured examples can create noise.
A useful sequence is often: clear example → contrasting example → near-miss → explicit statement of the invariant → fresh test case.
The goal is not volume. It is evidence about which features survive variation.
4. Concrete examples can support abstraction when they are well designed
Concrete contexts can make a new idea easier to enter because they connect to familiar knowledge.
But concrete features can also dominate attention. A colourful context may be remembered more strongly than the relationship it was meant to illustrate.
The correct conclusion is not “concrete is bad” or “abstract is always better.” The design question is whether the example helps the learner notice the target structure. A famous 2008 study reported an advantage for abstract examples in a particular mathematics task, while a later critical replication found that improved concrete examples performed as well as or better than the abstract condition. The debate itself is useful because it warns against universal prescriptions. Kaminski, Sloutsky and Heckler (2008); Trninic (2020).
5. Abstract symbols are not the same as abstract understanding
A page full of variables can look abstract while the learner is merely following symbol manipulation.
True abstraction is visible when the learner can explain what relationship the symbols preserve, why the representation is general, and which cases it does not cover.
Changing apples into x and oranges into y is not enough if the learner still cannot recognise the same relation in another situation.
6. Invariants are the core of useful abstraction
An invariant is something that remains structurally the same while other features change.
In equivalent fractions, the numerical appearance can change while value remains invariant. In an equation, equivalent transformations preserve equality. In a persuasive paragraph, topic may change while the claim–evidence–reasoning relation remains.
Ask learners explicitly: “What is allowed to change, and what must remain true?”
7. Abstraction must preserve conditions
Over-abstraction creates rules that are too broad.
A learner may abstract “larger denominator means smaller fraction” from examples with the same numerator. The rule works within that restricted family but fails generally.
A strong abstraction includes its boundary: larger denominator produces a smaller fraction when the positive numerator is held constant.
8. Non-examples protect abstraction from becoming vague
After stating a general rule, test a case that almost fits.
If the learner cannot explain why the near-miss fails, the abstraction may be verbal rather than operational. Boundaries sharpen what the general statement actually means.
9. Abstraction and schema formation are related but different
Schema formation organises knowledge into structured units. Abstraction identifies the more general relation that can organise multiple cases.
A learner may build a schema for solving percentage increase questions. Abstraction asks which relationship makes all those questions instances of percentage change in the first place.
10. Abstraction and generalisation are also different
Abstraction extracts a general relation. Generalisation uses that relation beyond the examples from which it was learned.
The sequence is often: compare cases → abstract structure → test structure on a new case → generalise if the boundary still holds.
11. Mathematics makes abstraction visible through representation
Consider three stories: a recipe doubles, a map scale enlarges, and a machine produces twice as many units per hour. The contexts differ. The multiplicative relation may be structurally similar.
Ask learners to express each in a table or equation, then identify which relation is invariant. The representation can make abstraction inspectable.
12. Science abstraction should stay answerable to mechanism
Several experiments may instantiate the same principle, but the learner should not generalise from superficial similarity.
For example, different cooling situations may share energy-transfer ideas while differing in material, geometry and environment. The abstraction should state the relevant mechanism and the conditions under which it applies.
13. English abstraction turns examples into reusable structures
A learner may study one strong introduction and copy its wording. Abstraction asks what the introduction is doing: establishing context, narrowing the issue, signalling position and preparing the reader for the argument.
The words can change. The rhetorical job remains.
14. Pattern recognition can precede abstraction—but pattern is not explanation
Learners often notice that certain features recur before they can explain why.
This can be productive. Repeated detection suggests a candidate invariant. But the candidate must be tested against examples and near-misses before it becomes a trustworthy abstraction.
“I keep seeing this shape” is a clue. “This relation is necessary because…” is a stronger model.
15. Good abstraction reduces memory burden without reducing meaning
A general rule can compress many examples into one organised relationship.
This is one reason abstraction supports expertise: fewer separate cases need to be treated as entirely new. But compression is only useful if the rule still predicts and explains correctly.
16. Abstraction should be reversible
A learner who truly understands an abstraction should be able to move back into examples.
Ask them to generate a valid example, a non-example and a boundary case. If they cannot instantiate the rule, the abstraction may have become empty language.
17. Abstraction becomes useful when it improves selection
The learner meets a new task and recognises the structure quickly enough to choose a relevant model or method.
This is the point at which abstraction begins to pay rent: it reduces the need to treat every new surface as a completely new problem.
18. The final receipt is transfer without boundary loss
A sound abstraction should travel farther than the original examples while still knowing where to stop.
Give a changed case. Ask which part of the old relation applies, which part does not, and what evidence supports that judgement.
What abstraction is not
- Abstraction is not replacing every concrete example with symbols.
- Concrete examples are not automatically inferior.
- A general statement without boundary conditions can be overgeneralised.
- Pattern recognition is not yet abstraction until the relation is explained and tested.
- Abstraction should be reversible into valid examples.
- The goal is transferable structure, not maximum removal of detail.
An abstraction diagnostic map
| What adults see | Possible abstraction issue | Useful next move |
|---|---|---|
| Solves recipe ratios but not map scales | Knowledge tied to surface context | Compare representations and identify invariant ratio |
| States broad rule that fails on edge cases | Over-abstraction | Add near-misses and boundary conditions |
| Manipulates symbols but cannot explain relation | Symbolic procedure without abstraction | Translate back to examples and meaning |
| Needs many examples but sees no common structure | Comparison not focused | Align cases feature by feature |
| Can state rule but cannot invent example | Verbal abstraction not operational | Generate example, non-example and changed case |
A practical abstraction cycle
- Choose two examples with the same deep structure.
- List what differs.
- Identify what remains relationally the same.
- State the invariant.
- Add a near-miss.
- Refine the boundary.
- Represent the relation more generally.
- Generate a new example.
- Test a changed context.
- Explain why the abstraction still applies—or why it stops.
For parents
- “What is the same between these two different-looking questions?”
- “Which detail could change without changing the method?”
- “What condition must stay true?”
- “Can you invent a new example of the same structure?”
- “Can you give me a near-miss where the rule stops working?”
For students
- Compare different surfaces deliberately.
- Write the relationship that remains invariant.
- Do not delete conditions from the rule.
- Generate examples and non-examples.
- Translate between concrete and symbolic forms.
- Test the abstraction on a fresh case before trusting it.
How do we know abstraction is improving?
- Learners identify deep similarities across different contexts.
- Surface features control method choice less often.
- General statements include better boundary conditions.
- Symbols are connected to meaning.
- New valid examples can be generated.
- Near-miss cases are rejected for the right reason.
- Transfer improves without false generalisation increasing.
The complete abstraction chain
COMPARE → ALIGN → REMOVE IRRELEVANT DETAIL → IDENTIFY INVARIANT → STATE BOUNDARY → REPRESENT GENERALLY → TEST → TRANSFER
Research and evidence boundary
Research on abstraction, analogical comparison and transfer supports the value of helping learners identify relational structure across examples, but it does not support a simple rule that abstract representations are always superior to concrete ones. Results depend on example design, learner knowledge, task and transfer test. Gentner, Loewenstein and Thompson found benefits from comparing cases for schema abstraction and transfer in negotiation learning, while the Kaminski–Trninic debate shows why claims about abstract versus concrete examples should remain bounded. Gentner, Loewenstein and Thompson (2003); Kaminski et al. (2008); Trninic (2020).
