Semantic priming—faster response to target words following related primes—reveals that activation spreads automatically through semantic networks. Priming effects depend on association strength and decay over time. This indicates semantic memory is organized as interconnected networks where activating one concept increases activation of related concepts.
The priming paradigm is deceptively simple: show a participant a word (the prime), then show them a target word and measure how quickly they can make a lexical decision (is this a real word?). When prime and target are semantically related — BREAD → BUTTER — responses are faster than when they are unrelated — NURSE → BUTTER. This semantic priming effect is typically 20–50ms, small but reliable, and it reveals something important about how knowledge is stored and accessed.
Spreading activation (Collins & Loftus, 1975) is the dominant explanation. In this model, semantic memory is organized as a network of nodes (concepts) connected by edges (associations). When you encounter BREAD, the BREAD node activates. That activation then spreads along edges to neighboring nodes — BUTTER, TOAST, CARBOHYDRATE, WHEAT — in proportion to the strength of the association. By the time you see BUTTER, its node is already partially activated, so the lexical decision process reaches threshold faster. Stronger associations produce larger priming effects; weaker or more indirect associations produce smaller effects that decay more quickly.
A critical feature of spreading activation is that it is automatic and passive — it does not require conscious attention or intention. Even if participants are told the prime is irrelevant to their task, priming effects still occur. This automaticity distinguishes semantic priming from strategic expectancy: if you see the prime DOCTOR and consciously predict that NURSE is coming next, you will show a priming effect even for non-associates — but this strategic effect takes longer to develop, is visible only at long prime-target intervals, and disappears when participants are told not to predict. The early, short-interval priming effect is purely the result of spreading activation through the semantic network.
The architecture of the network carries theoretical implications. Hub models (where abstract semantic concepts are stored in a central hub, such as anterior temporal lobe) and distributed models (where meaning emerges from patterns of activity across sensory-motor cortex) make different predictions about the structure of priming. Hub damage (semantic dementia) flattens the semantic network — priming effects degrade uniformly across all semantic categories. Sensory-motor area damage produces category-specific deficits: patients with motor cortex damage show reduced priming for action words; patients with visual cortex damage show reduced priming for visual-property words like "bright." Semantic priming thus provides a behavioral window into the organization of conceptual knowledge — not just whether two things are related, but how the mind represents the nature of that relationship.