A. Intra- and inter-sequence comparisons

Prior to build each experiment, a pre-test was conducted to control the temporal dynamics of each action sequence. Specifically, we controlled that, for any point within a given sequence, a particular type (‘transport’ vs. ‘lift’ vs. ‘rotate’) or model (transport[1] vs. transport[2], lift[1] vs. lift[2], etc.) of action was not more or less discriminable than another one. To do so, each action scene in each of the 4 experiments was first cut at {960, 1160, 1320, 1400, 1480, 1560, 1640, 1720, 1800, 1880, 1960, 2080} ms after the onset of the movement. The resulting movies (12 clips per model of action) were randomly presented to a group of 12 participants in four sessions. Each session consisted in a basic action recognition task. In the two first sessions, movies from the motor and superordinate experiments were presented with the instruction of recognizing the performed actions by pressing, as quickly and as accurately as possible, one of the corresponding keyboard buttons. In the two last sessions, movies from the social motor and social superordinate experiments were shown with the same instruction. The order of sessions was counterbalanced across participants. For the (non-social) superordinate and social (motor and superordinate) movies, the only last action was shown.

Two groups of statistical analyses were conducted on both hits and RTs.

First, we compared participants’ hits for each type of action within each session (‘lift’ vs. ‘rotate’ vs. ‘transport’ actions within non-social motor and superordinate experiments; ‘transport’ vs. ‘rotate’ actions within social motor and superordinate experiments), then we compared hits for actions of the same type between sessions (e.g. lift[motor] vs. lift[superord.]).

Two-tailed t-tests revealed no significant differences between participants’ correct responses for the different types of action within each session (comparing ‘transport’ vs. ‘lift’ vs. ‘rotate’ actions in non-social motor and superordinate conditions, all p>0.22; comparing ‘transport’ vs. ‘rotate’ actions in social motor and superordinate conditions, all p>0.4), as well as no significant differences for actions of the same type between sessions (non-social conditions: lift[m] vs. lift[s], rotate[m] vs. rotate[s], transport[m] vs. transport[s]: all p>.19; social conditions: transport[m] vs. transport[s], rotate [m] vs. rotate[s]: all p>.25).

Second, we were compared the individual distributions of hits and reaction times (RTs) across the different movie segments to ensure that participants’ performance was equally sensitive to variations in the amounts of visual information (see figure 9). To do so, we computed within each experiment a one-way Kruskall-Wallis test with participants’ hits and RTs as dependent variables and the amount of information as a 12-level factor.

As shown inFigure 10, the participants’ detection curve showed the usual sigmoidal shape, performance rapidly increasing within a 1400-1700 ms interval of movie durations up to the maximal value (100%). In all experiments, Kruskall-Wallis tests revealed no significant differences between individual distributions of performance, with, between participants, identically decreasing RTs (motor: H(11,144)=9.34, p=.59; superordinate: H(11,144)=7.64, p=.74; social motor: H(11,144)=6.34, p=.84; social superord.: H(11,144)=5.1, p=.89) and increasing rate of correct responses (motor: H(11,144)=2.88, p=.9; superordinate: H(11,144)=2.61, p=.9; social motor: H(11,144)=1.98, p=.9; social superord.: H(11,144)=1.02, p=.9) as the available amount of visual information increased.

Fig. 9.Distribution of participant’s reaction times (blue dots) across the 12 movie segments. Reaction times for the different actions were pooled across subjects. Red squares: mean reaction times across participants for each of the 12 duration ranges.