6.2 Concept Formation, Mental Representation, and Problem-Solving Strategies

Concept Formation, Mental Representation, and Problem-Solving Strategies

How does the human mind organize vast quantities of sensory and declarative information into coherent mental representations, and how are these representations transformed to solve novel problems? Cognitive psychology explores these foundational questions by investigating the structure of semantic concepts, the internal format of mental imagery, and the strategic search algorithms and heuristic shortcuts employed during problem solving.

1. Concept Formation and Categorical Representation

A concept is a fundamental mental representation of a class of objects, events, or abstract entities, whereas categorization is the cognitive process of assigning an entity to its appropriate conceptual grouping. Three primary theoretical frameworks describe how categories are mentally structured:

Classical / Definitional           Prototype Theory                  Exemplar Theory
(Aristotle, Frege)                (Eleanor Rosch, 1970s)             (Medin & Schaffer, 1978)
────────────────────────          ──────────────────────             ────────────────────────
- Necessary & sufficient rules   - Central summary 'average'        - Multiple concrete instances
- Clear-cut, binary boundaries    - Graded membership & typicality   - Retains category variance
- All members equal status        - Privileged 'Basic Level'         - Sensitive to context & atypicality

The Classical (Definitional) View and Its Failures

Originating in classical philosophy, the definitional view posits that concepts are represented as lists of singly necessary and jointly sufficient features (e.g., a bachelor is defined by [+male], [+adult], [+unmarried]; a triangle is defined by [+three-sided], [+closed plane figure]). An entity belongs to a category if and only if it satisfies every defining attribute.

• Fatal Theoretical and Empirical Flaws:
  1. Inability to Specify Defining Features: For common natural and cultural categories (e.g., game, fruit, art), theorists cannot identify necessary and sufficient criteria. Ludwig Wittgenstein (1953) noted that instances of games (board games, card games, Olympic sports, ball games) share no universal common denominator, but rather exhibit a complex network of overlapping similarities termed family resemblance.
  2. Fuzzy Category Boundaries: Category borders are often indefinite and probabilistic rather than discrete. People disagree on whether borderline items (e.g., rug as furniture, tomato as fruit) belong to a given category.
  3. Typicality Effects: Members of a category do not hold equal status. In speeded sentence verification tasks (e.g., "A robin is a bird" versus "A penguin is a bird"), participants respond significantly faster and with fewer errors to prototypical members than to atypical members, directly violating the classical assumption that all members meeting the definition are equivalent.

Prototype Theory (Eleanor Rosch)

Developed by Eleanor Rosch and colleagues in the 1970s, Prototype Theory asserts that categories are organized around a prototype—an abstracted, idealized central representation that reflects the statistical average or central tendency of characteristic features encountered across category members:

• Graded Structure and Typicality Gradients: Category membership is continuous rather than binary. The more characteristic features an exemplar shares with the prototype, the higher its typicality rating (e.g., robins and sparrows are highly typical birds; ostriches and penguins possess low typicality).
• Empirical Manifestations of Typicality:
  • Verification Speed: High-typicality items are verified faster in semantic sentence verification.
  • Production Frequency: When asked to generate exemplars of a category, high-typicality items are named first.
  • Priming Facilitation: Presenting the category name (e.g., "bird") primes responses to high-typicality exemplars more powerfully than to low-typicality exemplars.
• Levels of Categorization (Rosch's Taxonomy):
  • Superordinate Level: Broad, abstract groupings (e.g., Animal, Furniture, Vehicle). Low within-category similarity; members share few common perceptual attributes; low informativeness.
  • Basic Level (Cognitively Privileged): Intermediate groupings (e.g., Dog, Chair, Car). This level represents the optimal cognitive compromise between informativeness and distinctiveness. Members share a common global shape, evoke identical motor interaction programs, are learned first by young children, exhibit the fastest object identification latencies, and carry the highest cue validity.
  • Subordinate Level: Highly specific groupings (e.g., Golden Retriever, Windsor Chair, Sports Car). High within-category similarity, but low distinctiveness from neighboring subordinate categories.

Exemplar Theory (Medin & Schaffer, 1978; Nosofsky, 1986)

Exemplar Theory rejects the notion of an abstract, averaged prototype. Instead, it posits that a concept is represented by the aggregate collection of all specific, concrete individual exemplars previously encountered and stored in episodic/declarative memory:

• Classification Mechanism: When encountering a novel stimulus, the cognitive system computes its similarity against all stored exemplars of various categories. The stimulus is assigned to the category whose aggregate exemplar similarity is greatest.
• Advantages Over Prototype Theory: Exemplar theory readily accounts for human sensitivity to category variability, correlations among specific features (e.g., small birds sing, large birds do not), and the retention of atypical members (e.g., we recognize a penguin as a bird because we retrieve specific stored memories of penguins, not because a penguin resembles an abstract flying songbird prototype).

2. Mental Representation and Imagery: The Analog-Propositional Debate

How are perceptual experiences stored and manipulated in working memory in the absence of direct sensory input? This question sparked the famous analog versus propositional imagery debate:

           [ Analog / Depictive Code ]                         [ Propositional / Descriptive Code ]
             (Stephen Kosslyn, Roger Shepard)                     (Zenon Pylyshyn)
           ───────────────────────────────────                 ─────────────────────────────────────
           - Functional equivalence to perception              - Language-like, abstract symbols
           - Continuous, metric spatial coordinates            - Discrete predicate calculus: [ON(CAT, MAT)]
           - Directly preserves physical geometry              - Imagery is an epiphenomenal byproduct
           - Supported by mental rotation & scanning           - Tacit knowledge & experimenter demand

Dual-Coding Theory (Allan Paivio, 1971)

Allan Paivio proposed that long-term memory incorporates two separate, functionally independent yet interconnected representational subsystems:

1. Non-Verbal / Visual System (Imagens): Specialized for generating and manipulating analog mental images, processing information synchronously, spatially, and in parallel.
2. Verbal / Linguistic System (Logogens): Specialized for linguistic units, processing information sequentially, hierarchically, and propositionally.

• The Dual-Coding Advantage: Concrete words (e.g., piano, tiger) can be encoded simultaneously across both logogen and imagen systems, establishing two independent, redundant memory traces. In contrast, abstract words (e.g., justice, epistemology) are encoded almost exclusively in the verbal logogen system. Consequently, concrete words exhibit consistently superior recall, recognition, and retention (concreteness effect).

Empirical Hallmarks of Analog Representation

• Mental Rotation (Roger Shepard & Jacqueline Metzler, 1971): Participants viewed pairs of 2D perspective line drawings of 3D block figures and judged whether they were identical or mirror-image rotations. The angular disparity between objects varied from 0° to 180° in either the 2D picture plane or 3D depth plane.
  • Finding: Reaction time to confirm identity was a strictly linear monotonic function of the angular disparity between the two figures (~60 degrees per second).
  • Significance: Mental operations preserve physical spatiotemporal properties; subjects do not instantaneously compute propositional coordinate transforms, but mentally rotate representations smoothly through intermediate angular orientations in analog psychological space.
• Image Scanning (Stephen Kosslyn, 1973, 1978): In the famous fictional island map experiment, participants memorized an artificial island featuring seven distinct locations (e.g., hut, tree, well, beach) separated by varying metric distances. After removing the map, participants formed a mental image of the island and were instructed to focus on one landmark, mentally "scan" across to a target landmark, and press a button upon arrival.
  • Finding: Reaction time to scan between two imagined landmarks was a direct linear function of the physical distance separating them on the original physical map.
  • Mental Walk Paradigm: When imagining walking toward an animal, participants reported that the image filled the visual field at a distance proportional to the animal's physical size (visual angle overflow), mimicking perceptual optics.
• Pylyshyn's Propositional Counter-Critique: Zenon Pylyshyn (1973, 1981) argued that analog images are merely epiphenomenal—subjective experiential "byproducts" devoid of functional causal utility, akin to the heat generated by a light bulb. Pylyshyn contended that all mental representations are fundamentally abstract, discrete, language-like propositions, and that Kosslyn's mental scanning latencies reflected demand characteristics or subjects' tacit knowledge of physical laws rather than structural constraints of the cognitive architecture.
• Neuroimaging Resolution: Subsequent functional neuroimaging (PET and fMRI) strongly supported the depictive/analog model by revealing that visual mental imagery elicits retinotopic activation within primary visual cortex (V1 / Brodmann Area 17), with high-resolution imagery recruiting the central foveal representation and larger images extending into peripheral retinotopic fields.

3. Problem Solving: Frameworks, Algorithms, and Heuristics

A problem exists whenever an organism seeks to achieve a desired goal state but lacks an immediately obvious means of traversing the gap. Problems are categorized by their structural clarity:

• Well-Defined Problems: The initial state, allowable operators, and goal state are completely and unambiguously specified (e.g., chess, the Tower of Hanoi, algebraic equations).
• Ill-Defined Problems: The initial state, goal state, and/or valid operators are vague, ambiguous, or underspecified (e.g., selecting a career, writing an evocative poem, achieving subjective happiness).

Newell & Simon's Problem Space Theory

In their seminal work Human Problem Solving (1972), Allen Newell and Herbert Simon formalized problem solving as an information-processing search through a problem space:

[ Initial State ] ──(Operator A)──> [ Intermediate State 1 ]
                                           │
                                      (Operator B)
                                           │
                                           ▼
[ Goal State ] <───(Operator D)──── [ Intermediate State 2 ]

* Path Constraints: Resource limits, forbidden moves, and rules governing search.

1. Initial State: The baseline description of the problem at its outset.
2. Goal State: The target terminal configuration to be attained.
3. Operators: The permissible actions, transformations, or legal moves that transition the system from one state to another.
4. Intermediate Problem States: Every configuration generated between the initial and goal states.
5. Path Constraints: Rules or limitations that restrict legal operator execution (e.g., in the Tower of Hanoi, a larger disc may never be placed atop a smaller disc).

Algorithms vs. Heuristics in Problem Solving

• Algorithms: Exhaustive, systematic, step-by-step procedures that guarantee a correct solution if one exists (e.g., long division, testing every mathematical combination in a combination lock). In complex domains, algorithms suffer from combinatorial explosion, requiring computational resources that rapidly exceed human working memory capacity.
• Heuristics: Cognitive rules of thumb, informal strategies, or educated shortcuts that substantially reduce the search space within the problem space. They do not guarantee a correct solution, but provide rapid, computationally tractable solutions:

4. Obstacles to Problem Solving and the Nature of Insight

Human problem solvers frequently experience severe impasses caused by entrenched cognitive biases and structural inflexibility:

Cognitive Obstacles to Problem Solving

• Mental Set (Einstellung Effect): The tendency to persist in utilizing a previously successful problem-solving strategy or mental framework, even when the current problem can be solved by a much simpler, more direct method, or when the habitual method no longer works.
  • Luchins Water Jar Problems (Abraham Luchins, 1942): Participants were trained on a series of complex arithmetic problems requiring measuring specific water volumes using three jars ($B - A - 2C$). After repeated trials, participants were given an extinction problem easily solved by $A - C$ or $A + C$. Participants suffering from mental set blindly executed the convoluted $B - A - 2C$ algorithm, and many failed completely on problems where the old formula was invalid.
• Functional Fixedness: A specialized manifestation of mental set in which an individual's conceptualization of an object is rigidly confined to its typical, habitual function, blocking the realization that the object can be repurposed to serve a novel mechanical role.
  • Duncker's Candle Problem (Karl Duncker, 1945): Participants were provided a candle, a box of thumb-tacks, and matches, and asked to affix the candle to a cork wall so wax would not drip on the table. The solution requires emptying the box, treating it as a platform, tacking the box to the wall, and placing the candle inside. Robert Adamson (1952) demonstrated that when the box was presented full of tacks (highlighting its container function), fewer than half of participants solved the problem within the time limit; when the box was presented empty beside the tacks, most (about 86%) solved it.
• Irrelevant Information and Distraction: Problem solvers often assume that all numerical or physical details provided in a problem stem are necessary for its solution, cluttering working memory and diverting cognitive resources from core structural constraints.

Insight vs. Non-Insight Problem Solving

  Subjective Warmth Rating (1-10)
    10 ───┐                                 /  <─── Insight Problem (Sudden Jump)
          │                                / 
     7 ───┼                               │
          │             /─────────────────┘ 
     4 ───┼            / <───────────────────────── Non-Insight / Algebra (Linear Climb)
          │           /
     1 ───┴──────────/─────────────────────────────
          0%        25%         50%        75%      100%
                       Elapsed Time to Solution

• Gestalt Origins: Gestalt psychologists (Wolfgang Köhler, Karl Duncker, Max Wertheimer) argued that problem solving is not merely incremental trial-and-error associationism (as Edward Thorndike contended), but frequently involves insight—a sudden, holistic restructuring or perceptual reorganization of the problem space that illuminates the solution pathway (the Aha! experience). Köhler (1925) documented insight in chimpanzees (notably Sultan), who, after failing to reach suspended bananas, suddenly stacked packing crates or joined hollow bamboo sticks to secure the fruit.
• Metcalfe & Wiebe (1987) Warmth Ratings Experiment: Janet Metcalfe and David Wiebe empirically tested the psychological reality of insight by comparing participants solving routine / non-insight problems (e.g., multi-step algebraic equations) versus insight problems (e.g., the bronze coin problem, the nine-dot problem, water jar riddles). Every 15 seconds, participants reported their subjective "warmth" (closeness to the solution on a 1–10 scale):
  • Non-Insight Problems: Subjective warmth ratings increased gradually and incrementally in a steady linear slope as participants systematically executed intermediate computational steps.
  • Insight Problems: Subjective warmth ratings remained flat and low throughout the deliberation period, followed by a sudden, dramatic spike to maximum warmth immediately prior to achieving the correct solution.
• Neurocognitive Correlates of Insight: Mark Beeman, John Kounios, and colleagues (2004) recorded EEG and fMRI during Remote Associates Test (RAT) problem solving. They discovered a distinctive burst of high-frequency gamma-band oscillation (~40 Hz) over the right anterior superior temporal gyrus occurring approximately 300 ms prior to subjective insight, accompanied by sudden activation of the anterior cingulate cortex mediating cognitive restructuring.