The influence of shared visual context on performance with novel codes in a referential communication game

It has been proposed that ostensive communication enables complex languages to evolve (Scott-Phillips, 2014). Successful ostensive communication, both verbal and non-verbal, must rely on a body of shared information that has been described as “common ground” (Clark, 1996). By virtue of knowing similar things, communicators are better able to infer each other’s intentions. As a subset of the general phenomenon of common ground, shared visual context refers to the situation of two interlocutors building on visibly common information. The importance of shared visual context for referential communication has been extensively studied (e.g. Clark, Schreuder, & Buttrick, 1983; Hanna, Tanenhaus, & Trueswell, 2003), but its impact on language evolution remains an open question. Other studies (Winters, Kirby, & Smith, 2015; Tinits, Nölle, & Hartmann, 2017) have investigated the effects of different types of context on evolving languages, but not considered the shared aspect of this context. How important is shared visual context in the genesis of new languages through communication? We conducted two laboratory experiments to answer the question of whether shared visual context between two interlocutors could help solving the task of establishing novel codes, and using them for communication. In both experiments, dyads of participants played a game in which one subject was tasked to choose the target colour out of an array of four possible answers, while the second subject tried to communicate the target, using black-and-white abstract symbols only. The symbols represented abstract shapes and objects that are not limited to one natural colour – e.g., a diamond, a piece of candy, a butterfly. Shared visual context was manipulated between dyads by minimizing or maximizing what portion of the array seen by the receiver was also seen by the sender: In the absent visual context condition, the senders only knew about the target colour, whereas they had full access to the three distractors in the shared visual context condition. Crucially, participants never received feedback about correct or wrong answers during the game, but could only tell their partner whether they understood the message instead. This, together with the non332