Phil/Psych 256, Week 4

Concepts: Introduction

Discuss: What are some good examples of concepts?

Goals of a theory of concepts.

1. Philosophy: epistemology and metaphysics.

2. Psychology: understand thought.

3. AI: make powerful computational systems.

Review key issues about nature of concepts:

1. Psychological vs. nonnatural.

2. Innate versus learned.

3. Strictly defined vs. specifying typical conditions.

   Innate  Learned
 Typical  Fodor  Rosch
 Defined  Plato  Locke

Other possible views:

1. Concepts are purely linguistic (Wittgenstein).

2. Concepts don't exist: Skinner

3. Distributed representations (see Week 7).

4. Concepts as statistical associations: latent semantic analysis.

Concepts in philosophy:

Plato, Wittgenstein, issue of conceptual change.

Concepts in psychology:

Topic became respectable in 1950s. Rosch's experiments on concepts: some birds are more typical than others. Problem: some numbers are more typical than others.

Psychological theories of concepts:

1. Classical: definitions.

2. Prototype: Rosch.

3. Exemplar: store particular examples.

4. Causal: concepts used in explanation.

Computational theories of concepts:

1. Minsky's frames, 1975. E.g. room. A concept describes a typical situation that can be matched against reality. Cf. Schank's restaurant script.

2. Semantic networks: organizations of frames.

Representational power of concepts.

1. Concepts include basic information about a category, but are not intended to embrace all knowledge.

2. Problem: how much to include as part of a concept? I include some rules.

Concepts: Evaluation

Distinguish two questions:

1. What is the psychologically valid theory of concepts?

2. How important are concepts as components of mental representation?

Representational power of concepts

1. Concepts include basic information abnded to embrace all knowledge.

2. Problem: how much to include as part of a concept? I include some rules in my view of concepts. But concepts won't do as an entire representation scheme.

3. Semantic networks: power of hierarchical organization.

Computational power of concepts

Problem solving

1. Roles in many kinds of reasoning.

2. Different view of problem solving and explanation than what we have seen so far: matching, not deduction is crucial. E.g. apply frame (schema, script, prototype) of restaurant to explain why someone left money.

3. Note computational power of inheritance in semantic networks of frames (schemas, etc.).

4. Note also role of concepts in spreading activation. Associative problem solving.


1. Learning by definition: what is a concert?

2. Learning from examples.

positive and negative examples, e.g. arch.

3. Conceptual combination, e.g. apartment dog,

computational philosopher. Scientific discovery. Web potato.

Computational limitations.

1. Script based understanding programs turned out to be too limited: we don't have prototypes for everything. E.g. why is there a goat in the restaurant? Schank moved toward case-based (analogical) reasoning and abandoned natural language.

2. Minsky's frame idea got added into various expert systems, providing a supplement to rules, but some of his most suggestive ideas were never implemented.


Importance of the lexicon.

Compare WordNet: electronic thesaurus with nouns organized by kind and part relations.

Try out the Visual Thesaurus.

Key points in Medin

Phil/Psych 256

Computational Epistemology Laboratory.

Paul Thagard

This page updated Oct. 1, 2012