Concepts, Thought, and Rationality

Concepts are the reusable constituents through which subjects classify, infer, imagine, and think. The concept bird contributes to thoughts about birds across different contexts; the concept cause structures explanations; logical concepts support indefinitely many inferences. Theories disagree whether concepts are definitions, prototypes, exemplars, theories, abilities, or structured symbols.

No one account fits every concept equally well. Definitions explain exact logical and mathematical cases but struggle with ordinary categories; prototypes explain typicality but not compositionality; abilities fit embodied use but can leave truth conditions unclear. The relation between psychological processing and rational norms adds a second problem: how can finite, biased mechanisms instantiate standards of correct reasoning?

What concepts do

Concepts support:

• categorising new instances;
• drawing inferences;
• forming complex thoughts;
• recognising the same subject matter across presentations;
• planning and counterfactual reasoning;
• communicating through public expressions.

A theory can prioritise one role and explain the others. Prototype theories begin with classification, inferential theories with reasoning, and language-of-thought theories with compositional structure. The adequacy of the reduction depends on whether the remaining roles follow.

The classical definitional view

On the classical view, a concept is specified by necessary and sufficient conditions. A bachelor is an unmarried adult man; a triangle is a three-sided polygon. Categorisation checks whether an item satisfies the definition.

The view explains sharp boundaries, logical relations, and compositionality. It struggles to find definitions for ordinary concepts such as game, chair, knowledge, or species. People also judge some members more typical than others even when all satisfy the same conditions.

Definitions remain plausible for stipulated and formal concepts. Their limited range suggests conceptual pluralism rather than the disappearance of definitional structure.

Prototypes

A prototype represents a category through a central or average cluster of features. Items are classified by similarity, explaining why robins seem better examples of birds than penguins and why categorisation is graded.

Prototype models fit learning and speed. Their problems concern boundaries and composition. A pet fish is not well represented by combining the typical pet with the typical fish; constituent prototypes do not determine the prototype of the complex.

Similarity also depends on which dimensions receive weight. Without a task, theory, or goal, any two items resemble each other in indefinitely many respects.

Exemplars

Exemplar theories store particular encountered instances and classify new items by similarity to them. They preserve variability hidden by an average prototype and explain effects of frequency and recency.

The distinction between prototype and exemplar can be difficult to test because both produce graded judgments. Human cognition may use summaries and instances at different times.

Exemplars explain recognition better than abstract thought. A concept also supports counterfactuals and inferences about cases unlike those encountered, requiring theory or structure beyond stored similarity.

Theory theories

On a theory theory, concepts are nodes in bodies of causal and explanatory knowledge. The concept disease depends on views about causes, symptoms, transmission, and treatment rather than one feature list.

This explains why conceptual categorisation changes with background knowledge and why surface similarity can be overridden. It also creates holism: if theories differ, speakers may not share concepts, and radical conceptual change becomes hard to describe as change concerning one subject.

External reference can preserve continuity through theory change. Scientists before and after a discovery may refer to the same kind under different conceptions.

Concepts as abilities

An ability view identifies concept possession with capacities to discriminate, infer, imagine, act, and use expressions appropriately. Concepts are not inner objects but patterns of competence.

This fits embodied and pragmatic cognition and avoids asking where a concept token is stored. It can include non-linguistic agents whose practical abilities outstrip explicit definitions.

The view needs structure sufficient for novel composition. Possessing separate abilities for red and square should explain the capacity to think red square. Merely listing dispositions can reproduce the circularity problem faced by behaviourism.

Structured representations

Language-of-thought theories identify concepts with constituents of internal symbols. Their recombinability explains productivity and systematicity. The same red constituent appears in thoughts about red circles and red squares.

Atomic symbols need content. Causal, teleological, and inferential relations must connect the vehicle to redness. Primitive symbols may be too numerous if every lexical concept is innate, while learning new primitive concepts is difficult to explain without already possessing their constituents.

Distributed representations offer another format, as Connectionism explains. The key issue is whether constituent structure is explicit, emergent, or unnecessary for the target capacity.

Conceptual pluralism

Different tasks may recruit different representational structures. Categorisation can use prototypes and exemplars; explanation can use theory-like knowledge; formal reasoning can use structured symbols; skilled action can use embodied abilities.

Pluralism predicts context effects and uneven deficits. It risks replacing one theory with an unprincipled list. A mature account should specify how formats interact and why one is selected for a task.

Hybrid concepts may contain a lexical or referential anchor plus several bodies of information, none individually defining the concept.

Conceptual and non-conceptual content

Perception can discriminate more shades and shapes than a subject can classify. Infants and animals navigate spatial relations without the concepts expressed in adult language. These motivate non-conceptual content.

Conceptualists argue that perceptual experience can justify belief only if it already occupies the space of reasons. Non-conceptualists distinguish causal or representational content from the conceptual capacities needed to endorse a judgment.

The dispute partly concerns the standard for concept possession. If any systematic discrimination counts, non-conceptual content shrinks; if concepts require inferential and recombinable use, it expands.

Rational norms and psychological laws

Logic and epistemology specify how one ought to reason. Psychology describes how people actually reason, including heuristics, biases, and resource limits. A causal law cannot by itself establish a normative inference.

Yet rational norms must be usable by finite agents. A standard no possible reasoner could follow may fail as a cognitive ideal. Bounded rationality studies strategies that perform well under constraints rather than approximating unlimited deduction.

The relation can be layered:

• formal norms specify correctness;
• computational models specify procedures;
• mechanisms implement procedures;
• environments determine which procedures are ecologically effective.

No level reduces the others automatically.

Rational interpretation

Davidsonian interpretation attributes beliefs by finding a largely coherent pattern of truth and reason. A system wholly insensitive to implication or evidence may not count as holding propositional attitudes at all.

This makes rationality partly constitutive of content rather than a standard externally applied after attribution. It threatens to idealise real agents and to misclassify systematic irrationality as different meaning.

A modest version requires enough inferential integration for interpretation while allowing local error, conflict, and bias. Rationality is a background constraint, not perfect performance.

Conceptual change

Concepts change through discovery, social struggle, and practical innovation. Planet, gene, and marriage have altered in extension and associated theory. Change raises a continuity problem: if content is constituted by inferential role, how can one concept change rather than be replaced?

Causal reference, historical chains, and overlap in use can preserve topic. Sometimes replacement is the right verdict. The distinction is not purely linguistic; it affects whether past claims disagree with present ones or express different frameworks.

Conceptual engineering makes change deliberate, asking which concepts serve epistemic and social goals. Psychological tractability and normative improvement can pull in different directions.

Language and thought

Public language supports complex thought by supplying stable symbols, shared distinctions, and external memory. It may transform rather than merely express pre-linguistic cognition.

Thought is not universally linguistic. Spatial navigation, imagery, affect, motor planning, and animal cognition can proceed in other formats. Inner speech is one process among several and varies across people.

The question is therefore not whether language or thought comes first globally. It is which cognitive capacities require linguistic scaffolding and which contents can be represented independently.

Assessment

Concepts are likely heterogeneous cognitive tools rather than one representational kind. Rationality constrains their use without being reducible to actual dispositions. A good theory must connect normative structure with learnable, finite mechanisms rather than choose one side of that relation.

Selected references

• Carey, Susan. The Origin of Concepts (2009).
• Fodor, Jerry A. Concepts (1998).
• Machery, Edouard. Doing without Concepts (2009).
• Murphy, Gregory L. The Big Book of Concepts (2002).
• Peacocke, Christopher. A Study of Concepts (1992).