Simulation work in the field can be roughly divided into work on syntax (e.g., Batali, 1994, 1998; Briscoe, 1998; Gmytrasiewicz and Gopal, 2000; Hurford, 2000; Kirby, 1997, 2000; Kirby and Hurford, 2001; Steels, 1997; Tonkes and Wiles, 2002; to name but a fraction), the lexicon (e.g., Kaplan, 1998; Oliphant, 1996; Smith, 2001; Steels, 1998, Van Looveren, 2001; Vogt, 2001; as above) and the less common work on phonology/phonetics (e.g., de Boer, 2000, 2001; Glotin and Laboissiere, 1996; Berrah, 1998 (cited in de Boer, 2000)).
Though it would be impossible to give an accurate account of how each simulation was run, it is possible to review some of the main mechanisms. The basic format of many of the simulations (see Steels, 2000) can be described in the following way:
A group (sometimes as small as two) of agents is put together and engages in controlled interactions or language games. The agents usually have no ability to communicate initially, at least in the phenomenon being modelled and, after a certain number of language games, an ability and/or language develops with which enables the agents communicate. The extent of knowledge and abilities assumed varies (see below for discussion). The structure of the agents varies but they are commonly represented as neural networks (e.g., Batali, 1998; Tonkes and Wiles, 2002) or simply as a collection of numeric weights (de Boer, 2001) or physically instantiated as robots (Steels, Kaplan et al, 2002) with collections of numeric weights.
Steels (2000:1) describes the methodology thus:
The basic idea is that a community of language users (further called agents) can be viewed as a complex adaptive system which collectively solves the problem of developing a shared communication system. To do so, the community must reach an agreement on a repertoire of forms (a sound system in the case of spoken language), a repertoire of meanings (the conceptualisations of reality), and a repertoire of form-meaning pairs (the lexicon and grammar).
In the above quote Steels mentions the notion of a complex adaptive system. A full analysis of the notion of complex adaptive system is out of the question here and the reader is directed to Flake (2001) or Stonier and Xing (1995) for a more substantial treatment. The following excerpts give us an indication of what a complex adaptive system entails:
Complex Systems are things that consist of many similar and simple parts. Often the underlying behavior of any of the parts is easily understood, while the behavior of the system as a whole defies explanation. ... By changing the type and form or interactions that exist among the parts of a complex system, the type of global behavior can be varied such that the complex system as a whole can be globally goal-seeking while only local information is passed around by the parts. This means that a collective form of computation can take place without an explicit global algorithm. (Flake, 2001:229)
The analysis of systems with nonlinear interactions among system components dominates many aspects of current research. Such systems with interesting emergent behaviour are referred to as complex systems. Those complex systems with the additional property that their primitive components can change specification, or evolve, over time, are often called complex adaptive systems (CAS).
The basis of adaptation rests on the premise that there is some condition of operation or performance which is better than any other. Moreover, to be called adaptive, self-organising features must exist in the system to enable performance to be optimised. (Stonier and Xing, 1995:Introduction[2])
The basic idea is that groups of "agents" or units are designed to exhibit only very limited behaviours. Out of prolonged interaction a language or language ability develops.