Kinematic priming of action predictions

Abstract

The ability to anticipate what others will do next is crucial for navigating social, interactive environments. Here, we develop an experimental and analytical framework to measure the implicit readout of prospective intention information from movement kinematics. Using a primed action categorization task, we first demonstrate implicit access to intention information by establishing a novel form of priming, which we term kinematic priming: subtle differences in movement kinematics prime action prediction. Next, using data collected from the same participants in a forced-choice intention discrimination task 1 h later, we quantify single-trial intention readout-the amount of intention information read by individual perceivers in individual kinematic primes-and assess whether it can be used to predict the amount of kinematic priming. We demonstrate that the amount of kinematic priming, as indexed by both response times (RTs) and initial fixations to a given probe, is directly proportional to the amount of intention information read by the individual perceiver at the single-trial level. These results demonstrate that human perceivers have rapid, implicit access to intention information encoded in movement kinematics and highlight the potential of our approach to reveal the computations that permit the readout of this information with single-subject, single-trial resolution.

Keywords: intention; kinematic encoding; kinematic priming; kinematic readout; prospective information.

Figure 1

Experimental design and results of kinematic priming and intention discrimination (A) Experimental design and trial structure. Participants performed a primed action categorization task and, 1 h later, a forced-choice intention discrimination task. (B) Kinematic primes consisted of videos of reach-to-drink and reach-to-pour movements. (C) Time course of wrist height (W_H), wrist horizontal trajectory (W_HT), and grip aperture (G_A) for reach-to-drink and reach-to-pour movements. Thin curves show representative individual trajectories; thick curves with shaded areas show the mean ± SD across kinematic prime stimuli. (D) Response times (RTs) to action probes in the primed action categorization task by kinematic prime (reach-to-drink, reach-to-pour) and action probe (drinking, pouring). (E) Discrimination performance quantified as the predicted probability of correct choice in the intention discrimination task. In (D) and (E), the histograms show the estimated marginal mean ± SE at the population-level estimated from mixed models fit to single-trial data. See also Tables S1–S3 and Videos S1 and S2.

Figure 2

Kinematic coding framework (A) Schematic of the encoding (readout) model. (B) Model performance quantified as the predicted probability of correct choice by the encoding (readout) models. The histograms show the estimated marginal mean ± SE at the population-level estimated from mixed models fit to single-trial data. (C) Sketch of the kinematic encoding (left) and readout (middle and right) models in a simplified two-dimensional kinematic space. Elliptical regions represent the intention-conditional probability distributions in kinematic space. The encoding vector $\overrightarrow{{\beta }_{enc}}$ is the axis optimally discriminating reach-to-pour from reach-to-drink acts. The alignment of the readout vector $\overrightarrow{{\beta }_{read}}$(the axis used by perceivers to discriminate reach-to-pour from reach-to-grasp) relative to the encoding vector determines how efficiently the encoded information is read out. (D) Spearman correlation between intention readout and intention encoding at the single-prime level. For readout, data points represent the average intention readout across perceivers. The line and shaded region correspond to estimated marginal mean ± SE estimated from a linear model fit to individual prime data. (E) Color map plotting the contribution of individual kinematic variables to encoding (top) and readout of individual perceivers (bottom, 1–20). (F) Probability of confidence rating in the intention discrimination task as a function of single-trial intention readout. As single-trial intention readout increases, the probability of a lower confidence rating decreases and the probability of a higher confidence rating increases. Lines and shaded regions correspond to estimated marginal means ± SE estimated from the cumulative link mixed model. See also Figure S2 and Tables S1–S3.

Figure 3

Single-trial intention readout predicts RTs to action probes (A) RTs to action probes by single-trial intention encoding and congruency. (B) RTs by single-trial intention readout and congruency. (C) Priming effect (incongruent-congruent) by low and high encoding information. (D) Priming effect (incongruent-congruent) by low and high readout information. Mean lines, shaded areas and histograms and error bars represent estimated marginal means ± SE at the population level. Dotted vertical lines (A) and (B) indicate the median value of single-trial intention encoding/readout used for median-split in (C) and (D). See also Figures S3–S5 and Tables S1–S3.

Figure 4

Kinematic priming of initial fixations (A) Quadrants of task-relevant information in action probes. The shaded area indicates the quadrant of each probe containing task-relevant information. (B) Probability of initial fixation to the quadrant relevant for the displayed probe by kinematic prime (reach-to-drink, reach-to-pour) and action probe (drinking, pouring). (C) Probability of initial fixation to the quadrant relevant for the non-displayed probe by kinematic prime (reach-to-drink, reach-to-pour) and action probe (drinking, pouring). Histograms plot the estimated marginal mean ± SE at the population-level estimated from mixed models fit to single-trial data. See also Figure S4 and Tables S1–S3.

Figure 5

Single-trial intention readout predicts initial fixations (A) Probability of initial fixation to the region containing the most task-relevant information by single-trial intention encoding. (B) Probability of initial fixation to the region containing the most task-relevant information by single-trial intention readout. (C) Priming effect (incongruent-congruent) by low and high encoding information. (D) Priming effect (incongruent-congruent) by low and high readout information. Conventions are as in Figure 3. See also Tables S1–S3.