Schilbach et al. make a powerful case that researchers should study social cognition and social neuroscience using interactive rather than observational paradigms. In this commentary, we claim that the most important basis for interaction is communication (where the interlocutors know that interaction is necessary for them to achieve their goal). As communication typically involves language, it is especially important that the language sciences assume a second-person perspective.

Perhaps surprisingly, the vast majority of work on language has considered isolated examples of production (e.g., naming pictures) or comprehension (e.g., reading texts), whether using behavioral or neuroscientific methods (e.g., Traxler Reference Traxler2012). But it is undeniable that dialogue is the basic site for language use, in both developmental and evolutionary terms (Clark Reference Clark1996; Pickering & Garrod Reference Pickering and Garrod2004). It is therefore the right time to conduct extensive “second-person” research in language use.

The little experimental work on dialogue has tended to involve two interlocutors jointly performing a task (e.g., Clark & Wilkes-Gibbs Reference Clark and Wilkes-Gibbs1986) and indicates the importance of feedback on communicative success (e.g., Bavelas et al. Reference Bavelas, Coates and Johnson2000). There is clear evidence of entrainment, at levels from syntax (Branigan et al. Reference Branigan, Pickering and Cleland2000) to speech rate (Giles et al. Reference Giles, Coupland and Coupland1991). Recent work has studied brain activation during speaking and listening (Menenti et al. Reference Menenti, Gierhan, Segaert and Hagoort2011) and showed that communicative success correlates with the extent to which activity in the listener's brain anticipates activity in corresponding areas of the speaker's brain (Stephens et al. Reference Stephens, Silbert and Hasson2010).

But we believe that greater understanding is likely to result from studies of moment-by-moment processing in pairs of participants. Pickering and Garrod (Reference Pickering and Garrod2103) have argued that highly interactive uses of language in dialogue are rendered possible by underlying (predictive) mechanisms common to language production and comprehension. To test accounts such as this, we propose the use of joint language tasks in a way that is analogous to testing whether co-actors form representations of each other's actions (i.e., of their own performed actions and of another's observed actions; Sebanz et al. Reference Sebanz, Bekkering and Knoblich2006). Such tasks should of course involve two speakers “talking together”; however, we argue they need not necessarily involve fully-fledged interaction (i.e., “talking to each other”).

One way of “talking together,” for example, is talking at the same time. Albeit rare in natural conversations, it is theoretically interesting to investigate what happens when production and comprehension occur simultaneously. If the two share mechanisms, one would sometimes expect interference when they are engaged concurrently (Gambi & Pickering Reference Gambi and Pickering2011). Another way of “talking together” is completing each other's utterances. Paradigms in which speakers are asked to jointly produce a sentence allow tests of the hypothesis that, given shared mechanisms between production and comprehension, coordinating with others is similar to coordinating with oneself. For example, timing in speech is affected by properties of upcoming linguistic elements (e.g., Griffin Reference Griffin2003). One could therefore ask whether it is also affected by the properties of an expected continuation by somebody other than the current speaker.

Although such tasks are unlike natural dialogue in many ways, they allow us to test the role of predictions about what another speaker is about to say. Thus, they can investigate questions like: How do beliefs about one's interlocutor influence anticipatory adaptation to their upcoming utterances? What is the minimum degree of interaction necessary for such beliefs to be taken into account in the first place? To what extent are beliefs modified as a result of the interaction? In addition, such tasks allow tight experimental control and therefore permit a meaningful comparison between “isolated talking” and “talking together,” since in natural conversations it is much harder to disentangle the relative contribution of beliefs about one's partner from the direct influence of what they say or do during the interaction itself.

The latter point also relates to the interesting observation that there are various ways of entering a joint activity, with explicit commitments to shared goals and seamlessly automatic entrainment playing different roles at different times (Tollefsen & Dale Reference Tollefsen and Dale2012). Interestingly, the gaze coordination task proposed by Schilbach et al. could be used both in the investigation of how low-level entrainment influences the perception of being in a joint activity and in the study of how higher-level beliefs influence interactional dynamics (sect. 3.2.1). Similarly, with the joint sentence production task mentioned above, it would be possible to investigate how the degree of achieved coordination influences beliefs and feelings towards one's interlocutor, as well as the extent to which previous knowledge about one's interlocutor and their utterances affects the amount of coordination attained.

Finally, communication is best seen as existing on a continuum from monologue at one end to truly interactive dialogue at the other. In a casual conversation between intimates, the “flow” from one speaker to the other is likely to be seamless and the conversation is internally managed (i.e., the participants control the nature of the interaction). In a more formal situation such as an interview, a large-group discussion, or an audience attending to a story, the opportunities to interact are limited and are constrained by social rules.

Linguistic communication, therefore, represents an ideal case for the study of what it means to interact, because it allows researchers to compare behavior or neural activity in settings involving different degrees of interactivity. The study of communication is therefore likely to be very fruitful in the development of a second-person perspective in the cognitive sciences and neurosciences.