The cultural approach to language is ripe, let us expand our options to test it empirically. Commentary on "Modeling the cultural evolution of language" by Luc Steels.

For almost two decades now, Luc Steels has been a leading pioneer of two novel approaches to studying language evolution. Methodologically, he has pioneered the artificial approach [1, henceforth AA]; theoretically, he has pioneered the cultural approach [2, henceforth CA]. In this paper [3], Steels (i) situates these approaches within their broader scientific context, (ii) provides an integrated overview of the main contributions made by these approaches, and (iii) suggests a number of fundamental challenges for the CA. Since my comment concerning (i) and (ii) is extremely terse—Steels offers us a very timely and authoritative review—I will focus on (iii). Insightfully summarizing the key phenomena linguists have observed in natural languages, Steels identifies three challenges for the CA. The CA “should explain how a language system may emerge and continue to change”, it should “explain how a new language strategy can emerge and propagate in a population”, and it “should explain the semiotic dynamics we see in cultural language evolution”. Steels then describes how these challenges have been tackled empirically by researchers who adopt the AA, providing a number of exciting insights. Here I describe another empirical approach to tackle the challenges identified by Steels and suggest that this approach, which has been labeled Experimental Semiotics [4], provides a key methodological complement to the AA. In the last few years, researchers have begun studying languages that human adults develop de novo under controlled laboratory conditions [5–7]. To do this, experimental semioticians ask people to play cooperative games, typically in virtual environments. The games require communication but players cannot use pre-established forms of communication such as spoken or written languages. In order to succeed, players must develop novel languages. Most people do this relatively rapidly, providing new opportunities of investigation for students of language. Some of these opportunities parallel the opportunities offered by the AA, allowing us to test its findings with real human agents [e.g., 8,9]. However, I will not focus on these opportunities here. Rather, I will focus on an opportunity which does not parallel those offered by the AA, therefore providing a methodological complement to it. When faced with ES tasks, people sometime violate our expectations about their behavior. These violations are extremely valuable because they