A semantic network represents knowledge as a web of interconnected concepts. Each concept is a node (BIRD, ROBIN, CANARY, FEATHERS, FLY), and the relationships between concepts are labeled links (ROBIN is-a BIRD; BIRD has FEATHERS; BIRD can FLY). First proposed by Quillian (1967) and elaborated by Collins and Quillian (1969), semantic networks formalized the intuition that human knowledge is organized as an associative structure in which the meaning of any concept is defined by its relationships to other concepts. This representation has influenced cognitive psychology, artificial intelligence, and our understanding of how meaning is stored and retrieved from long-term memory.
Key Structures
- Temporal lobe — The brain region critical for auditory processing, language comprehension, memory formation, and object recognition — bridging perception with meaning.
- Frontal lobe — The largest lobe of the cerebral cortex, responsible for executive functions including planning, decision-making, working memory, and the voluntary control of behavior.
- Insight — The sudden, conscious realization of the solution to a problem — the 'aha!' or 'eureka' moment — often preceded by an impasse and accompanied by a feeling of certainty and surprise.
- Recognition — A form of memory retrieval in which a previously encountered item is identified as familiar when presented again, typically easier than recall because the target item itself serves as a retrieval cue.
- Long-Term Memory — The vast, relatively permanent storage system that holds knowledge, experiences, skills, and facts for periods ranging from minutes to a lifetime.
- Spreading Activation — The process by which activating one concept in a semantic network automatically sends activation to related concepts, facilitating their retrieval — the mechanism underlying priming, association, and .
The Hierarchical Model
Collins and Quillian's original model proposed that semantic networks are organized hierarchically, with properties stored at the most general applicable level to minimize redundancy. The property "has feathers" is stored at the BIRD node (not separately at ROBIN, CANARY, and SPARROW), and "breathes" is stored at ANIMAL (not at BIRD or FISH). Retrieval involves traversing links: verifying "A canary has feathers" requires moving from CANARY to BIRD and finding "has feathers," while "A canary breathes" requires moving from CANARY to BIRD to ANIMAL. This hierarchical storage predicts that verification time increases with the number of links traversed, a prediction confirmed in many experiments.
Revisions and Spreading Activation
The strictly hierarchical model encountered problems: "A robin is a bird" is verified faster than "A penguin is a bird," even though both require the same single link traversal. Collins and Loftus (1975) proposed a revised model replacing strict hierarchy with spreading activation. In this model, activating a concept sends activation spreading along all connected links, with activation decreasing with distance and link strength. Concepts that are more closely associated (more frequently co-occurring, more similar, more strongly linked) activate each other more quickly. This model explains typicality effects (robin activates BIRD faster than penguin does) and semantic priming (hearing "doctor" speeds recognition of "nurse").
Neuroimaging studies have provided evidence that semantic knowledge is organized in the brain in ways that resemble semantic networks. Concepts from the same category (animals, tools) activate overlapping brain regions, and the temporal lobe's anterior regions appear to function as a "hub" connecting distributed semantic representations across modality-specific cortical areas. Semantic dementia — progressive degeneration of the anterior temporal lobes — produces a pattern of knowledge loss consistent with degraded network structure: patients lose the ability to distinguish between similar concepts before losing broader categorical knowledge, as if fine-grained links deteriorate before coarse ones.
Legacy and Modern Developments
Semantic networks have had an enduring influence. In artificial intelligence, knowledge graphs (such as those powering search engines) are direct descendants of Quillian's semantic networks. In cognitive psychology, the network metaphor continues to shape theories of memory retrieval, semantic priming, and knowledge organization. Modern approaches have enriched the basic framework: feature-based models represent concepts as collections of weighted features, embodied approaches add sensory-motor grounding, and distributional models derive semantic similarity from statistical patterns of word co-occurrence in text. Yet the core insight — that meaning resides in relationships between concepts, not in isolated definitions — remains as valid as ever.
Disorders
- Disrupted semantic networks in semantic dementia
- altered network structure in schizophrenia (loose associations)