Source
Zacks et al. (2007), LIT-1382, read in full in NOTE-1084 from the author manuscript in PubMed Central (PMC2852534). The theory is in “A Theory of Event Segmentation”; the evidence in the two sections that follow.
What was actually shown
The paper states a mechanism and assembles evidence for its implications.
- The mechanism. “Perceptual systems continuously make predictions about what will happen next. When transient errors in predictions arise, an event boundary is perceived” (Abstract). Event models are “working memory representations”; updating “consists of (1) resetting the current representations … and (2) transiently increasing the influence of the pathway from sensory inputs to the event models”. It is claimed to run on several timescales at once.
- What could have come out otherwise. A neural network trained to predict an actor’s body positions (Reynolds, Zacks & Braver, then in press) might not have shown larger prediction error at event boundaries, or gained from error-gated event models; it did. Readers who met a time shift (“an hour later” rather than “a moment later”) might have recognized what came just before as well as controls; they did not (Speer & Zacks 2005). Boundaries marked later might have drawn no transient brain response during earlier passive viewing; they did, in extrastriate and right frontal cortex. And the anterior cingulate might have responded at boundaries during passive viewing, as the theory predicts; it did not, which the authors call the one prediction that “does not square well with the available data”.
- Supporting regularities. Unitization is reliable, hierarchical across grains, and tied to movement change, goal completion and sequential statistics. Boundaries are remembered better, and segmenting well predicts later memory.
What this does not say
- That prediction error has been shown to cause boundaries in people. It has not been measured. Change, goals and statistics each predict boundaries; error is the theory’s proposed common cause.
- Anything about one’s own intentions. “A second limitation is that the theory in its current form is a purely passive perceiver.” Extending event models to represent “one’s own goals” is named as future work. An intention lost at a boundary (CASE-136, CLAIM-421) is an extension of the theory, not something it states.
- The doorway effect. The paper does not mention it. Its memory prediction is that what preceded a boundary becomes harder to retrieve. The doorway studies find the cost on the object carried across the boundary, which they explain by interference between event models (THEORY-888), a different mechanism.
- Where it lives in the brain. The mapping to cingulate, midbrain and prefrontal cortex is offered as tentative, and part of it failed.
- Consciousness. Event models “are not necessarily accessible to consciousness”. It shares a prediction-error architecture with THEORY-883 but makes no claim about experience.
Connections
- CASE-136: the case calls this “the theory behind” the doorway effect. It is the theory behind the event-cognition framing, not the account its authors give of the effect.
- THEORY-888: the doorway effect explained as interference between event models; compatible with segmentation, but not derived from this reset mechanism.
- TERM-055: EST’s split between event models (reset at boundaries) and event schemata (long-term) is one mechanism for holding knowledge that is not to hand.
- THEORY-883: shares the prediction-error architecture; applies it to conscious access rather than to segmentation.