“IF a piece is pinned to my king, THEN do not move that piece.” That’s the one stated as a rule of behavior, not a fact — and the test for which bucket something belongs in is exactly that phrasing test: can you say it as “X is true” (declarative), or does it only make sense as “when situation Y, do Z” (procedural)?
- “The knight on d5 is pinned” — a fact about the current board. Declarative: a chunk, true right now, retrievable, statable, forgettable.
- “A knight moves in an L-shape” — also a fact, just a permanent one rather than a transient one about this game. Still declarative: you can state the rule of the game even though you don’t use it as a trigger-condition the way you use the pin-avoidance habit.
- “IF pinned THEN don’t move it” — this is the production. Notice it isn’t “a fact I know about pins” — it’s a disposition to act a certain way whenever the triggering pattern shows up in declarative memory. You could, if pressed, articulate it as a sentence (experts often can, when teaching) — but the psychologically real claim ACT-R makes is that the behavior runs off a compiled production whether or not you can narrate it, which is why skilled players “just see” the right move faster than they can explain why.
- “My opponent is rated 2400” — a fact, declarative, exactly like the first.
Why the split earns its keep, beyond taxonomy: declarative and procedural memory predict different failure modes and different learning curves, and both show up in real players. Declarative retrieval can fail outright (you blank on an opponent’s name) or come back wrong (misremember a rating) — those are activation failures, this lesson’s number faltering. A production, once compiled by practice, rarely “forgets” that way; instead it can misfire — apply in a context that looks similar but isn’t (a pin-avoidance habit triggering on a piece that only looks pinned) — a different error signature entirely, because it’s a pattern-match failure, not a retrieval failure.
The other side of the split, seen from lesson 3: a production system’s IF–THEN rules were already procedural memory, just without the declarative half named yet. ACT-R’s move was making explicit that a production’s IF-side matches against declarative content (chunks in working memory) — the recognize–act cycle, now with its two halves named and each given its own retrieval dynamics (chunks decay and get boosted by activation; productions get faster and more selective with practice, previewed in stage 4’s power-law-of-practice lesson).
Where this goes: activation was defined here as “a number capturing retrievability” without saying where the number comes from. Next: what actually drives activation up or down — spreading activation from related chunks, and decay from disuse — and the genuinely surprising ACT-R claim that forgetting is the rational response to how memory gets used, not a bug.