Predictive Processing in the Context of Action Perception

Le résumé fourni par la source

Perception involves the continuous generation and testing of predictions about incoming sensory information (Friston, 2005; Rao & Ballard, 1999). The brain generates expectations about forthcoming events based on prior knowledge, and these predictions are compared against actual sensory input. When predictions match the input, processing proceeds efficiently, and the perceptual representation is accepted with little additional processing. When there is a mismatch, prediction errors signal the need to update internal models (Alexander & Brown, 2018; Clark, 2013). These errors are passed through the processing hierarchy, triggering a reassessment of the initial prediction and the generation of revised expectations that better account for the incoming sensory data. The prediction process is iterative where comparison and revision continue until the discrepancy between expected and observed input is reduced to an accepted level. Consequently, processing is slower when predictions are violated, as additional computational steps are required to resolve the mismatch. This framework extends to action observation, where observers generate predictions about others' behavior based on contextual cues about their goals and intentions (Kilner et al., 2007). The perception of physical objects’ motion is formed from variables such as direction and velocity, but human action is formed by perceived internal states goals and intentions within a given context (Bach et al., 2014; Hudson, Nicholson, Simpson, et al., 2016). Predicting what another person will do therefore requires sensitivity to cues that signal these intentions. Hudson, Nicholson, Ellis, et al. (2016) tested this using representational momentum – the tendency to misremember moving objects as further along their trajectory than they actually were (Freyd & Finke, 1984). Participants heard an actor state an intention to take or leave an object, then watched the actor reach toward or withdraw from it. The remembered final hand position was displaced further forward when the action matched the stated intention than when it contradicted it, suggesting that congruent intention-action pairings facilitate rapid acceptance of the predicted trajectory. Intention statements are not the only source of such predictions. Object properties generate similar effects: a hand approaching a graspable object is perceived as further along than one moving toward empty space (Hudson, Nicholson, Simpson, et al., 2016). Observers also exploit situational constraints and kinematics (e.g., deceleration patterns, grip aperture) to anticipate upcoming action (Stapel et al., 2012). Action prediction thus integrates explicit statements, object affordances, context, and movement dynamics into a unified perceptual expectation. Representational momentum provides a useful index of the perceptual process. Because the magnitude of forward displacement reflects how readily observers extrapolate a trajectory into the future, it captures the degree to which a prediction (extrapolation) has been accepted at the moment an action disappears. Larger displacement indicates rapid integration of the predicted trajectory, whereas smaller displacement suggests that ongoing reassessment has generated revised expectations. Findings indicate that representational momentum for observed human action results not only from extracting low-level motion information to anticipate future positions but is also influenced by top-down information from prior knowledge about the actor's goals. According to predictive processing models, perceptual representations emerge from a Bayesian-like integration of what was predicted and what was observed, with greater weight placed on expectations when the input is ambiguous (Clark, 2013; Kok et al., 2013). When observing brief action sequences that are subject to uncertainty, top-down predictions exert a stronger influence on perception, biasing it toward expected outcomes. Hubbard (2019) proposes a hierarchical account of these effects. At one level, displacement reflects broadly learned regularities of the environment - basic physical constraints such as momentum and gravity that the perceptual system routinely anticipates. On top of this baseline, stimulus and situation- specific knowledge can modulate the direction and magnitude of displacement, including object identity, expected changes in behaviour, and the observer's beliefs about what is happening. On this view, representational momentum is often automatic, but it is not fixed: what an observer expects can shift how far they "carry" the action forward in memory. Importantly, these expectations are themselves shaped by learning history. With repeated exposure to particular action-outcome contingencies, observers can become more likely to anticipate certain continuations, and this can strengthen (or reduce) the bias in remembered position. This aligns with predictive processing accounts in which perceptual representations reflect an integration of prior expectations and incoming sensory evidence (Clark, 2013; Kok et al., 2013). When the sensory information is brief, noisy, or ambiguous, top-down predictions should exert a stronger influence, biasing perception toward the most likely outcome given the observer's prior knowledge and the current context. In our previous unpublished research, we extended these findings using food-related actions paired with hunger and non-hunger intention statements. Congruent intention-action pairings elicited larger representational momentum effects than incongruent pairings. We also observed block effects: congruent trials produced larger representational momentum in earlier blocks, when the reliability of the intention cue was presumably highest following an initial priming block of congruent-only trials. These findings align with predictive coding accounts proposing that cue reliability modulates the weighting of prior expectations in perceptual inference. However, these findings do not establish whether congruent intention information increases the representational momentum effect above baseline levels, incongruent information suppresses it below baseline, or both processes occur. Comparing congruent and incongruent conditions establishes that they differ but does not reveal whether this difference arises from facilitation, disruption, or both. The current experiment (Experiment 3) aims to address this by introducing a neutral control condition in which participants observe hand movements without goal objects or intention statements. Moving stimuli produce representational momentum due to bottom-up extrapolation of motion dynamics (Freyd & Finke, 1984; Hubbard, 2005, 2019). When top-down expectations from intention statements are absent, the resulting representational momentum should reflect this baseline effect driven primarily by the observed kinematics. The neutral condition therefore provides a reference point against which to evaluate the influence of intention cues. If congruent intention statements facilitate predictive processing, the congruent condition should exceed the neutral baseline. Participants will observe hand movements towards or away from food objects. In congruent trials, the intention statement (hunger or non-hunger) predicts the subsequent hand movement (reach or withdraw) towards goal objects (food objects). In neutral (control) trials, no intention statement precedes the hand movement and there are no goal objects. On each trial, participants will see six rapid successive presentations of a hand that imply consistent leftward or rightward motion. After the hand disappears, a probe stimulus will appear showing the hand in one of five positions relative to the actual disappearance point: -24, -12, 0, +12, +24 pixels. Participants will indicate whether the probe shows the hand in the same position as when it disappeared or in a different position. The weighted mean measure will serve as the primary index of represe

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.