How does movement kinematics combine with other sources of information in revealing others’ intentions? There are situations in which the intention of an observed actor can be unambiguously estimated from one source of information, e.g., the type of grasp, the presence of a target object. Most often, however, combining different sources of information may lead to more accurate predictions. This is indeed what Stapel et al. (2012) demonstrated by asking participants to anticipate how an observed action would unfold. Participants observed an actor walking. After a few steps, they had to indicate how the action would continue, i.e., whether the actor would take another step walking or start crawling. A first experiment showed that observers were more accurate when they could base their predictions on the combination of movement kinematics, situational constraints (e.g., the presence of a table), and target object position (a ball). In a second experiment, the target object was artificially moved to another location so that movement kinematics was incongruent with the target object position. Results revealed that, in this ambiguous situation, participants relied on movement kinematics rather than on object location in making their predictions. This suggests that in the presence of conflicting information from different sources, movement kinematics may be prioritized to disambiguate the agent’s intention. A challenge for future research will be to understand the temporal course of information integration from different sources. A recent transcranial magnetic stimulation (TMS) study by Cavallo et al. (2013) demonstrated that, at movement onset, motor-evoked potential responses reflected the most probable motor program estimated from the situational context (e.g., whole hand grasp). During movement observation, however, the initial motor program was substituted by a new plan matching the specific features of the observed movement (e.g., precision grip). Thus, an intriguing possibility is that the contribution of movement kinematics is related to the specific stage of the observed action processing: before the to-be-observed action starts, observers rely on contextual factors to predict the course of the action; as the movement unfolds, however, action prediction might prioritize kinematic information. If confirmed, this would have implications for the interpretation of the so-called chain model of action organization (Bonini et al., 2013): modulation of mirror neuron discharge by end-goal might reflect not only (and not so much) the presence of contextual cues allowing the monkey to predict the experimenter’s intention (Fogassi et al., 2005), but also sensitivity to intention-related differences in the movement kinematics.

The studies reviewed above used spatial and temporal occlusion procedures to quantify pick-up of advance information. The advantage of using psychophysical methods is the high degree of control and statistical power they ensure. However, it is not clear how far this type of paradigm accounts for real-time interactions in which two or more individuals are set in a common social context. Social cognition has been proposed to be substantially different when we actively interact with others (“second-person” social cognition) rather than merely observe them (“third-person” social cognition; Schilbach et al., 2013). For third-person social cognition, observing body movement is merely a way of gathering data about the other person. For second-person social cognition, the knowledge of the other resides – at least in part – in the interaction dynamics “between” the agents (De Jaegher et al., 2010); it is thus plausible that interaction dynamics affect pick-up and use of advance kinematic information.

An initial investigation on this topic was made by Streuber et al. (2011) by adapting the spatial occlusion procedure to a social interaction task. Participants played a table tennis game in a dark room with only the table, the net, and the ball visible. The game could be played in a cooperative fashion, i.e., to play the ball back and forth as often as possible, or in a competitive fashion, i.e., to win the trial. The visibility of the players’ racquets and the body movements was manipulated with the following logic. If a specific source of information is important for playing table tennis, then rendering this source of information visible should positively affect the players’ performance. Results revealed that when the game was played cooperatively, seeing the other player’s racket had the largest effects on performance. In contrast, when the game was played competitively, seeing the other player’s body resulted in the largest increase in performance. This suggests that online cooperative and competitive dynamics selectively modulates the use of visual information about others’ actions. A question to be addressed by future research is whether a similar modulation is observed in offline tasks, in which participants are required to merely observe cooperative and competitive actions. More generally, it would be interesting to directly compare second-person and third-person social understanding with respect to the pick-up and the use of advance information: is attunement to kinematic features modulated by self-involvement? Do second-person and third-person intention understanding rely on the same kinematic characteristics?

Ever since their discovery, mirror neurons have been proposed to underlie our ability to understand actions “transforming visual information into knowledge” about others’ goals and intentions (Gallese and Goldman, 1998). But how exactly is this transformation achieved?

Rizzolatti and Craighero (2004) suggested a rather simple mechanism: “Each time an individual sees an action done by another individual, neurons that represent that action are activated in the observer’s premotor cortex.” This motor representation of the observed action “corresponds to that which is spontaneously generated during active action and whose outcome is known to the acting individual.” In this way, mirror neurons would transform visual information into knowledge about another person’s intention.

This model has been criticized on the assumption that “the same visual kinematics can be caused by different goals and intentions” (Kilner et al., 2007). Simulating the observed kinematics – it has been claimed – might allow an observer to represent what the agent is doing. However, given the non-specificity of the observed kinematics, it will not allow them to represent the agent’s intention (Jacob and Jeannerod, 2005).

The findings reviewed above provide strong evidence to the contrary. First, in contrast to the “non-specificity assumption,” they demonstrate that intention information is specified in the visual kinematics. Second, they indicate that observers are sensitive to this information and can use it to discriminate between different intentions. Evidence that the mirror system supports this ability comes from recent fMRI studies (Vingerhoets et al., 2010; Becchio et al., 2012). For example, Becchio et al. (2012) report that mirror areas are sensitive to kinematic cues to social intention. Participants observed isolated reach-to-grasp movements performed with the intent to cooperate, compete, or perform an individual movement, followed by a static test picture. They were required to judge whether the test picture depicted a continuation of the observed movement or not. Despite the lack of contextual information, observing grasping movements performed with a social intent relative to grasping movements performed with an individual intent activated mirror areas, including the inferior frontal gyrus and the inferior parietal lobule. Interestingly, comparison of social vs. individual movements also revealed differential activations at the temporo-parietal junction and within the dorsal medial prefrontal cortex, two regions traditionally associated with explicitly thinking about the state of minds of other individuals (i.e., “mentalizing”). These findings shed some light on the neural mechanisms underlying intention-from-movement understanding. They leave, however, a number of crucial issues unanswered.

A first issue pertains to how observed actions are mapped onto one’s own motor system. The mirror system is generally assumed to associate observed actions with “corresponding” motor programs of the observer. What though is exactly meant by “corresponding?” When we observe other individuals act, the very fact that our body differs from theirs’ introduces a disparity between the observed and the executed kinematics (for data on this issue see for instance Gazzola et al., 2007). It is thus difficult to envision how, at a computational level, the executed kinematics might be “coupled” with the observed kinematics (but see Press et al., 2011).

A second question concerns the exact contribution provided by the mirror and the mentalizing system (Van Overwalle and Baetens, 2009). While some theorists have argued that these two systems are mutually independent (e.g., Jacob and Jeannerod, 2005; Saxe, 2005), a substantial number of authors support the notion that the mirror system might inform the mentalizing system (e.g., Keysers and Gazzola, 2007; Uddin et al., 2007). According to this view, people would use their own motor system to encode the intentionality of an action based on its visual properties and form a pre-reflective representation of the other person’sintention. This representation would then serve as inputs to attributional processing within the mentalizing system (Keysers and Gazzola, 2007; see also Spunt and Lieberman, 2012). In line with this, de Lange et al. (2008) report that mirror areas, including the inferior frontal gyrus, process the intentionality of an observed action on the basis of the visual properties of the action, irrespective of whether the subject paid attention to the intention or not. In contrast, brain areas that are part of the mentalizing network become active when subjects reflect about the intentionality of an observed action, but are largely insensitive to the visual properties of the observed action. Alternatively, mirror neurons might discharge during action observation not because they are driven by the visual input but because they are part of a generative model that is predicting the sensory input (Kilner, 2011). Within this framework, the generative model starts with a prior prediction of the intention of the observed action. This prediction would be estimated in areas outside the mirror system (including mentalizing areas) and then conveyed to mirror areas, influencing the selection of a specific action intention. Techniques for characterizing effective connectivity between brain areas can provide answers in this debate because they can demonstrate the influence one system exerts over the other.