Semantic networks

Quillian represented a word's meaning as a node in a graph linked to other nodes, and showed how to find what two concepts share.

Why it matters

Knowledge was first represented as a structure a machine traverses rather than a list of statements, the ancestor of knowledge graphs.

The model found an intersection by spreading out from two nodes at once, and explained why people connect nearby concepts more easily. Minsky's frames, Schank's scripts and description logics descend from it. Present-day knowledge graphs keep the same structure of nodes and typed edges. Collins and Quillian tested the model in 1969 with a true-false reaction-time task: if a fact such as that a canary can fly is stored only at the node for bird, deciding it should take longer than deciding that a canary can sing, which is stored at the node for canary; by their abstract the results supported that organisation.

Event record

Event date

October 1966 – 1968

Timeline date

Event date

Verification

Sources gathered automatically · September 17, 2026

Lines

Symbolic AI

ID

evt-0112

The BBN report is dated October 1966; the text was reissued in Minsky's 1968 collection. The anchor is the collection.

Sources

primary Semantic Memory
Bolt Beranek and Newman · Published October 1966
primary A. M. Collins, M. R. Quillian, Retrieval Time from Semantic Memory, Journal of Verbal Learning and Verbal Behavior 8, 1969, pp. 240-247
Academic Press (scan hosted on a university teaching page) · Published 1969

Related events

Earlier

Related Syntactic Structures 1957
The model describes meaning through links between concepts rather than syntactic rules.

Later

Related Frames 1974links to this record
A frame adds to a network of concepts a structure of expectations with fillable slots.
Related Scripts, Plans, Goals and Understanding 1977links to this record
Meaning comes from a structure of expectations, not only from links between concepts.
Extends Bayesian networks 1985links to this record
A graph of concepts acquires probabilistic semantics and a rule for propagating evidence.
- Bayesian Networks: A Model of Self-Activated Memory for Evidential Reasoning · UCLA / Cognitive Science Society
Related WordNet: a dictionary arranged by meanings 1990links to this record
The same shape as Quillian’s semantic networks: node, arc, node. The specification does not cite them - the words semantic net occur zero times in it - so this is a reading alongside, not a claim of descent.
Related PageRank and web scale 1998links to this record
A node's importance is derived from link structure, as in semantic networks, but at web scale.
Related RDF: a fact as a triple with a global address 1999links to this record
The same shape as Quillian’s semantic networks: node, arc, node. The specification does not cite them - the words semantic net occur zero times in it - so this is a reading alongside, not a claim of descent.