Modeling the Evolution of Communication: From Stimulus Associations to Grounded Symbolic Associations
چکیده
This paper describes a model for the evolution of communication systems using simple syntactic rules, such as word combinations. It also focuses on the distinction between simple word-object associations and symbolic relationships. The simulation method combines the use of neural networks and genetic algorithms. The behavioral task is influenced by Savage-Rumbaugh & Rumbaugh’s (1978) ape language experiments. The results show that languages that use combination of words (e.g. verb-object rule) can emerge by autoorganization and cultural transmission. Neural networks are tested to see if evolved languages are based on symbol acquisition. The implications of this model for Deacon’s (1997) hypothesis on the role of symbolic acquisition for the origin of language are discussed. 1. Symbol acquisition in the evolution of communication The synthetic approach of Artificial Life has recently been applied to studying the evolution of communication and language (Steels, 1997). Some models have been used for the simulation of the emergence of simple lexicons in populations of simulated organisms (e.g. Cangelosi & Parisi, 1998; MacLennan & Burghardt, 1994) or in small communities of robots (Steel & Vogt, 1997). In these studies organisms evolve shared lexicons for describing entities and relations of the environment. Other models have focused on the evolution of syntax (e.g. Kirby & Hurford, 1997). Simulated organisms evolve different syntactic languages starting from a given set of syntactic structures and constraints, and devices for syntax acquisition. The first type of model, that focus on lexicon emergence, do not make any explicit reference to the role of syntax in language origin. Their aim is to model the early stages of the evolution of (animal) communication. Indeed, in animal communication systems, no syntactic structures have been observed. For example, no animal communication systems have been found that share one of the main properties of human languages, i.e. the combination of words to express different and new meanings. These models of lexicon evolution study communication systems based on simple signal-object associations. Organisms learn and evolve simple stimulus associations between objects in the environment and signals. In the second type of language origin models, researchers have taken the availability of some form of syntactic structures for granted. Their models do not explain how syntax can emerge and the role of syntax for organisms’ adaptation and survival. The associations that organisms learn are self-referential symbol-symbol relationships. These models are subject to the symbol grounding problem (Harnad, 1990) since they lack an intrinsic link between their symbols and the entities and relations existing in the organisms’ environment. Internal symbols need some form of sensorimotor grounding. Due to the symbol grounding problem, the role of these models for understanding the evolution of cognition is reduced. Recently, Terrence Deacon (1997) proposed an explanation for the fact that animal communication and human language differ. A variety of animal communication systems have been studied (Hauser, 1996), however, there is no apparent continuity between animal communication systems and complex human languages. That is, no “simple languages”, using some elementary forms of word combinations or syntax, have been found in the animal kingdom. The existence of simple languages could explain the big gap between animal and human communication. Deacon (1997; 1996) believes that this is due to the symbol acquisition problem. In fact the main difference between animals and humans relies on symbolic references. There is a significant difference between the referencing system of simple object-word associations and that of symbolic associations. In animals, simple associations between world entities and words can be explained by mere mechanisms of rote learning and conditional learning. An animal acquires genetically, or learns, that a word’s sound is always associated with a specific object. Instead, symbolic associations have double references, one between the word (symbol) and the object, and the second between the symbol itself and other symbols. A language-speaking human knows that a word refers to an object and also that the same word has logical (syntactic) relation with other words. Due to the possible combinatorial interrelationships between words, there can be an exponential growth of reference with each new added word. The difference between these types of associations, and their relation to the models of language origin, is graphically represented in Figure 1. Figure 1a represents a communication system based on simple associations between objects and words. It refers to the models of the acquisition of lexicons. Figure 1b represents the models of the origin of syntax. It only shows word-word associations, but this system is not based on real symbolic associations as a link is missing between words and objects. Words are self-referencing and they lack a grounding in the external world. Figure 1c shows a system based on grounded symbolic associations. The arrows represent references between words, and references between words and objects. This is the type of system that we will study in this paper. It should be noted that the relationship between words and objects, that constitute the grounding of symbols to entities of the real world, is not a direct link between mental symbols and real objects. Instead, it is a link between mental entities (the symbols or words) and other mental entities (such as concepts) that constitute the semantic reference. These categorical representations, that Deacon (1996) calls “indexical”, are useful to “sort out” the extensive perceptual variability of objects in the real worlds. The ability of humans and animals to create mental categories, e.g. through categorical perception, constitutes the “groundwork” of cognition (Harnad, 1987). From this it is possible to build more complex cognitive skills, such as language. “juice” “pour” “give"