Emergence of Natural Language Lexicons: Empirical and Modeling Evidence from Homesign and Nicaraguan Sign Language

Where do we get language from? Clearly, it requires our human brains: a chimp exposed to a lifetime’s worth of language will not surpass a child who has only had a few years of exposure. But it also clearly requires something from the environment: language-deprived children clearly don’t spring forth speaking Hebrew, or Greek, or Sanskrit. What is it about the human learner, and about the environments (environment broadly construed, as we will see) that they are typically exposed to, then, that give rise to language? The answer, put deceptively simply, is that language emerges when many individual humans interact; that is, language self-organizes in a population. In this paper, we consider how conventional lexicons self-organize within populations. We present data from two classes of naturally emerging signed communication systems, and we use an agent-based computational model (that is, a model that captures individuals, their interactions, and their emergent behavior) to explore the role that social networks play in the rates of conventionalization. Language emergence has thus far mostly been studied with computational, or, to a lesser extent, experimental methodologies (Experimental Semiotics). The experimental work brings human participants into the lab to accomplish some task in pairs or groups, deprived of familiar channels of communication (speech, writing, gesture). Thus, participants must create a new communication system. The systems that emerge in these settings are surprisingly language-like, possessing conventions (Galantucci, 2005), compositionality at the sublexical (Galantucci, Kroos, & Rhodes, 2010) and lexical levels (Selten & Warglien, 2007), and form-meaning mappings that become more arbitrary with use (Theisen, Oberlander, & Kirby, 2010). While examining different aspects of