The Pearce-Hall model (1980) addresses a limitation of the Rescorla-Wagner model by allowing the learning rate (associability, α) of a conditioned stimulus to change dynamically based on recent prediction errors. When outcomes are surprising (large prediction error), associability increases; when outcomes are well-predicted, associability decreases.
The Updating Rule
αₙ₊₁ = γ·|λₙ − ΣVₙ| + (1−γ)·αₙ₋₁
S = salience of the CS (fixed)
α = associability (variable, changes with surprise)
γ = weighting of current vs. previous associability
Key Predictions
The model explains latent inhibition (pre-exposure to a CS without the US reduces α, slowing subsequent learning), the Hall-Pearce effect (partial reinforcement maintains high α, facilitating later acquisition), and the finding that surprising events capture attention. Neurally, the associability signal has been linked to acetylcholine release from the basal forebrain — surprising events increase cholinergic activity, enhancing cortical plasticity and attention.