Timing of grip and goal activation during action perception: a priming study

Abstract

Several models of action recognition acknowledge the involvement of distinct grip and goal representations in the processing of others’ actions. Yet, their functional role and temporal organization are still debated. The present priming study aimed at evaluating the relative timing of grip and goal activation during the processing of photographs of object-directed actions. Action could be correct or incorrect owing to grip and/or goal violations. Twenty-eight (Experiment 1) and 25 (Experiment 2) healthy adults judged the correctness of target actions according to object typical use. Target pictures were primed by action pictures sharing the same grip or same goal, both the same grip and same goal or none. Primes were presented for 66 or 300 ms in Experiment 1 and for 120 or 220 ms in Experiment 2. In Experiment 1, facilitative priming effects were observed for goal and grip similarity after 300 ms primes but only for goal after 66 ms primes. In Experiment 2, facilitative priming effects were found for both goal and grip similarity from 120 ms of prime processing. In addition, results from a control condition in Experiment 2 indicated that mere object priming could partially account for goal similarity priming effects, suggesting that object identity may help the observer to make predictions about possible action goals. Findings demonstrate an early and first activation of goal representations, as compared to grip representations, in action decoding, consistent with predictive accounts of action understanding. Future studies should determine to what extent the timing of grip and goal activation is context-sensitive.

Appendices

Appendix 1: List of objects used in Experiments 1 and 2

Carafe

Coffee cup

Fork

Hairbrush

Hairdryer

Hammer

Knife

Liquid soap

Magnifying glass

Pencil

Phone

Screwdriver

Tea spoon

Teapot

Toothbrush

Torch

Cream Tube

Water bottle

Watering can

Wine glass

Appendix 2

Priming paradigms are known to be influenced by low-level visual features (e.g., Kristjansson 2008). Thus, our results may be driven by perceptual differences between conditions and not by the activation of different levels of action representation. For example, earlier and stronger priming effects may be expected between pairs of action pictures that are more similar perceptually. We used the FSIM algorithm developed by Zhang et al. (2011) to assess image similarity based on low-level visual features. The closer to 1 the index is the more pictures are similar perceptually. First, an index of perceptual similarity was computed between each type of prime (correct grip but incorrect goal, incorrect grip but correct goal, incorrect grip and incorrect goal, neutral action-free) and target (correct action target, incorrect action target) and then attributed to each type of prime-target pairs (Grip similar only, goal similar only, all different, neutral). No indices were computed for grip and goal similar pairs, as prime and target were the same exact picture.

Perceptual similarity was modelled as a function of prime-target pairs (Grip similar only, goal similar only, all different, neutral) and RESP (yes, no) as fixed effects and items as random intercepts. Model comparison did not show any interaction between RESP and type of pair [χ²(3) = 2.5356, p = .4689] and no main effect of RESP [χ²(1) = 1.7477, p = .1862]. We did observe, however, a main effect of type of pair [χ²(3) = 213.11, p < .001]. Compared to ‘all different’ pairs, perceptual similarity was higher in ‘similar grip only’ pairs (estimate 0.055, SE = 0.004, t = 12.36, p < .001), but lower in neutral pairs (estimate − 0.044, SE = 0.004, t = − 9.96, p < .001). ‘All different’ and ‘goal similar only’ pairs were not different from each other (estimate 0.007, SE = 0.004, t = 1.67, p = .09).

To assess the relation between grip and goal priming effects and the perceptual similarity indices in the different goal and grip similarity conditions, Spearman’s rank correlations were computed. Table 2 summarizes the results of the analysis.

Table 2 Spearman’s rank correlations between goal and grip similarity priming effects and the corresponding perceptual similarity difference scores

Appendix 3A: Determination of the random effect structure in mixed-effect models of Experiment 1

Following the recommendation of Barr et al. (2013), we first attempted to fit a model with the maximum random structure, including, for both subjects and items, random intercepts, random slopes for GRIP, GOAL, DURATION, and RESP and random slopes for the four possible interactions between GRIP, GOAL, DURATION, and RESP. As expected, the model was overparametrized in regards to the data available and failed to converge to a stable solution (see Bates et al. 2015a, b; Matuschek et al. 2017). Then, following the latest recommendations of Bates et al. (2015a, b) we determined the optimal random structure supported by the data. First higher order random slopes (random slopes reflecting interactions) were removed. Then we looked for possible further reduction of the remaining random effect structure by conducting a principal component analysis on the random terms of the model using the rePCA function from the RePsychLing package version 0.0.4 developed by Bates et al. (2015a, b). The analysis identified 5 components, three of them being sufficient to explain 100% of variance. The random slope for DURATION was the least representative factor of these three components for subject and was thus removed from the random effect structure of the model. For items, the least representative factor was GRIP and GOAL, and the two factors were then removed from the random effect structure of the model. Consequently, the final model included random intercepts, GRIP, GOAL and RESP as random slopes for subject, and random intercepts, DURATION and RESP as random slopes for items.

The following full model was finally considered in Experiment 1 (bold indicates fixed effects of interest)

Appendix 3B: Determination of random effect structure in mixed-effect models of Experiment 2

As for Experiment 1, the maximum random structure was not suited for the data of Experiment 2. Moreover, the full model did not converge with GRIP, GOAL, DURATION, and RESP as random slopes for participants and items. A first principal component analysis on the random terms of the model indicated that the GRIP, GOAL and DURATION for subjects were the least representative factors of the components explaining 100% of variance and were removed from the analysis. For items, the least representative factors explaining 100% of variance were GRIP and GOAL and were then removed from the analysis. Thus, the final model included random intercepts and random slopes for RESP for subjects, and random intercepts, RESP and DURATION as random slopes for items.

The following full model was finally considered in Experiment 2 (Bold indicates fixed effects of interest)