Source Summary: Minsky 1974 — A Framework for Representing Knowledge
Full title: A Framework for Representing Knowledge
Authors: Marvin Minsky
Affiliation: MIT Artificial Intelligence Laboratory
Venue: MIT-AI Laboratory Memo 306, June 1974; reprinted in The Psychology of Computer Vision, P. Winston (ed.), McGraw-Hill, 1975
PDF: papers/minsky-1974-a-framework-for-representing-knowledge.pdf
Contributions
- Frames — a data structure for representing stereotyped situations; the foundational concept for structured knowledge representation.
- Frame systems — collections of related frames linked together; shifts between frames mirror changes in situation understanding.
- Default values and inheritance — slots have expected values that apply unless overridden by specific information.
- Unified theory of memory, perception, and language — frames apply to visual scenes, stories, conversations, and conceptual understanding.
The Core Idea (from PDF)
“When one encounters a new situation (or makes a substantial change in one’s view of the present problem) one selects from memory a structure called a Frame. This is a remembered framework to be adapted to fit reality by changing details as necessary.”
A frame is a data-structure with:
- Top levels: fixed slots representing what is always true about the situation (the type of frame)
- Lower levels: terminals (slots) that must be filled with specific values
- Default values: what to assume when a slot is not explicitly filled
- Conditions: constraints on what can fill a slot
Frame Structure
[Birthday Party Frame]
location: (default: someone's home)
guests: (must be: persons)
activity: (default: cake, presents, singing)
host: (must be: person)
duration: (default: 2–3 hours)
When you hear “birthday party,” you retrieve this frame. Unknown details are filled with defaults. New information overrides defaults.
Frame Systems
Related frames are linked into frame systems. Example: a room viewed from different angles — each viewpoint is a separate frame, but they share terminals and are linked by transformation rules.
Shifting between frames = updating your model of the situation:
- Enter “office building lobby” → retrieve lobby frame
- Move to elevator → transition to elevator frame (inheriting building context)
Why Frames Matter
Prior knowledge representations stored isolated facts (predicate calculus) or simple associations. Frames addressed the poverty of atomic representation:
- Context: a frame bundles all expected properties of a situation together
- Defaults: unknown information is assumed typical, not unknown
- Inheritance: properties shared across related frames don’t need to be re-stated
- Procedural attachment: slots can have attached procedures (demons) that fire when the slot is filled
Influence
| Descendant | Connection |
|---|---|
| Object-oriented programming | Classes = frames; instances = filled frames; inheritance = frame inheritance |
| Semantic networks | Frames as nodes with labeled relations |
| KL-ONE (1977) | Formal frame-based knowledge representation |
| Scripts (Schank & Abelson 1977) | Frames for event sequences |
| HTML/DOM | Structured document as a frame with slots |
| Modern ontologies (OWL) | Frame logic formalized |
Key References (from paper)
- Newell & Simon 1972 — Problem spaces (related structured representation)
- Schank 1974 — Conceptual Dependency (related event representation)
- Papert & Minsky 1972 — Micro-worlds