Analogy in Reasoning Ability
Advanced University‑Level Notes (MIT‑style)
1. Introduction
An analogy is a comparison that highlights a correspondence (a mapping) between two entities, sets of entities, or systems. In reasoning ability tests, analogies assess how well a test‑taker can uncover structural similarities that are not immediately obvious from surface features.
Analogy as a Cognitive Tool
- Analogy is a foundational mechanism in human cognition, used for learning, problem solving, and scientific discovery.
- In formal logic and artificial intelligence, analogy is formalized as relational mapping and structural equivalence.
These notes present the theory from first principles, derive rigorous criteria for validity, discuss edge cases, and culminate in a challenging problem that tests deep understanding.
2. Formal Foundations
2.1 Basic Definitions
| Concept | Symbol | Description |
|---|---|---|
| Term | X, Y, … |
An atomic element (word, concept, object). |
| Relation | R |
A binary or n‑ary predicate linking terms. |
| Mapping | M : S → T |
Function that associates each term in source set S to a term in target set T. |
| Structural Equivalence | ≡ |
S ≡ T iff there exists a bijective mapping M preserving all relations. |
Analogy Sentence: A : B :: C : D
Interpretation: “A is to B as C is to D.”
2.2 Relational Mapping Principle
Principle:
An analogy is valid iff there exists a bijective mapping M from the source pair (A, B) to the target pair (C, D) that preserves all explicit relations.
Formally: [ \forall R; (R(A,B) \implies R(M(A), M(B))) \quad\text{and}\quad \forall R; (R(C,D) \implies R(M^{-1}(C), M^{-1}(D))). ]
Why bijection?
- Guarantees one‑to‑one correspondence.
- Avoids many‑to‑one or one‑to‑many pitfalls that lead to false analogies.
2.3 Example (Simple)
- Source:
Water : Liquid :: Ice : Solid
Mapping:Water ↔ Ice,Liquid ↔ Solid.
All relations (phase, temperature, state) are preserved.
3. Advanced Theory
3.1 Structural Similarity Metrics
For complex systems, exact bijection is often infeasible. We use similarity function σ:
[ σ(S,T) = \frac{|R_S ∩ R_T|}{|R_S ∪ R_T|} ]
where R_X is the set of all relations in system X.
σ = 1→ perfect structural equivalence.σ < 1→ partial similarity; may still be a good analogy if key relations dominate.
3.2 Shepard’s Universal Law of Generalization
Empirical observation:
[
P(\text{recognize } \text{B as similar to } \text{A}) \propto e^{-k d(A,B)}
]
where d(A,B) is the metric distance in conceptual space.
Implication: Analogies are more likely when source and target lie close in semantic space.
3.3 Relational Entailment
An analogy can entail new relations:
-
Transitive Entailment
IfArelates toBbyR1andCrelates toDbyR1, thenAshould relate toCby a derived relationR2. -
Analogical Inference
FromA : B :: C : D, inferA : C :: B : Donly if mapping preserves the direction of relations.
3.4 Edge Cases & Pitfalls
| Case | Example | Why It Fails |
|---|---|---|
| Circular Analogy | A : B :: B : A |
Trivial mapping; no new insight. |
| Over‑extension | Sun : Star :: Earth : Planet |
Earth is a planet, not a star; mapping violates type compatibility. |
| Compositional Mismatch | Bird : Fly :: Fish : Swim |
Both are animals, but the action relation differs in context (air vs. water). |
| Non‑bijective Mapping | A : B :: C : D with A ↔ {C,D} |
Many‑to‑one mapping; loses clarity. |
| Ambiguous Relations | Tree : Branch :: River : Current |
“Branch” can mean branch of a tree or branch of a river. |
4. Practical Guidelines for Solving Analogy Questions
-
Identify Core Relations
- List all explicit relations (
R1, R2, …) between the first pair. - Exclude trivial or incidental relations (e.g., color unless explicitly stated).
- List all explicit relations (
-
Eliminate Surface‑Feature Matches
- Surface similarity