Article Preview
Top1. Introduction
Recently, a range of approaches has been proposed to address the emergence of linguistic categories shared among members of the same language community (Puglisi et al., 2008; Baronchelli et al., 2010). What these approaches have in common is that they start from a situation in which the community does not share any relation of expression and meaning units to end up with a common (form-)meaning function. This goes back to Luc Steels’ seminal work on the naming game as a simulation model of language evolution (Steels, 1996). Recently, it has been reconstructed in terms of self-organizing complex dynamical systems (Baronchelli et al., 2006; Dall’Asta et al., 2006a, 2006b). In this framework, the meaning function – as a result of successful language learning – consists of a single tuple that relates a finally unique expression unit with a semantic singleton. The reason to simplify the model in this way is to systematically study the impact of a large set of parameters while keeping the complexity of the model manageable. During simulation, the semantic singleton s formally diversifies in relation to several, synonymous expression units, but is finally expressed by a single expression provided that the naming game is successful. As a matter of fact, the ontological provenance of s is disregarded – it is just a placeholder of what is regarded in opposition to its alternative formal manifestations (see Figure 1).
Figure 1. Semantic diversification and semantic unification
The naming game scenario has been further developed to capture meaning functions of (non-compound) “lexical” items to (compound) semantic units as a result of articulating (Hjelmslev, 1969) a reference-semantic continuum (Figure 1). In this sense, categorization means the shared articulation of an amorph substance, shared among the members of the corresponding community (Puglisi et al., 2008; Baronchelli et al., 2010). As a touchstone of this Extended Naming Game (ENG), color terms are made an object of simulations that remarkably reconstruct several of their linguistic characteristics. In this sense, the ENG is outstanding as its predictions are in accordance with linguistic hypotheses.
In spite of this success, several questions are raised that regard the linguistic status of color predicates in particular and of linguistic predicates in general (Murphy, 2002; Taylor, 2003). The reason is that the ENG basically deals with non-compound predicates (e.g., red), which are directly related to their meanings regardless of the formation of compound signs (e.g., red bowl). This reduction leaves out all cases in which lexical units (e.g., adjectives and nouns) are combined to express complex categories based on more elementary ones. As a matter of fact, this includes non-lexicalized categories for which complex expressions are needed to express them, that is, the majority of categories being manifested by natural languages. Thus, the ENG currently disregards semantic compositionality and contextuality, that is, the systematic relation of the meaning of compound units, their constituents and contexts.
At first glance, our assessment of the ENG seems to be pedantic as it abstracts from the complexity under consideration to keep manageability of its parameter space.1 However, the question arises how to extend any such framework to cope with more realistic categorization scenarios in accordance with the complexity of natural language. In other words: how shall we extend the complexity of our model step-by-step to capture which aspect of the complexity of linguistic categorization? Or to put it more concrete: How shall we extend the ENG to capture semantic compositionality and contextuality?