Over the last decade a brand of nativism has taken hold of behavioural science in the form of Evolutionary Psychology. Advocates of this theoretical approach argue that peripheral and central cognition is organised as a modular architecture, each module reflecting a solution to an ancestral problem from the Pleistocene (Cosmides and Tooby, 1992; Pinker, 1997). This Evolutionary Nativism is at odds with recent ontogenetic work that argues for a canalized development whereby constraints guide and shape a more general associative learning system to achieve specific ends (Elman, Bates, Johnson, Karmiloff-Smith, Parisi and Plunkett, 1996; Karmiloff-Smith, 1992). As such one can talk about emergent modularization where ontogenetic development leads to an adult end-state that is module like. This recent perspective has the potential to alter current modes of adaptationist thinking within psychology. In order to explore this potential a theoretical model of symbol origins is proposed that relies on specific constraints canalizing expectancy and operant learning. It is proposed that this system of constraints underwent Baldwinian selection thereby instantiating, in modern Homo sapiens sapiens, the ability to learn symbols rapidly. This accords with specific ontogenetic theories of word acquisition (Bloom and Markson, 1998, Markson and Bloom, 1997).

In this paper I assume that symbols are arbitrarily and symmetrically related to their referent, and in this way symbols are afforded the property of displaced reference. The relation is arbitrary because the symbol has no direct causal relationship with the referent. For example, there is nothing "rabbit-like" about the word <rabbit>. <Rabbit> is attached to rabbits by social convention (Deacon, 1997). Furthermore, if someone points to an array of objects and says <rabbit> we can pick out the rabbit from this array. Equally, we can say <rabbit> when someone presents us with a rabbit. In this way the symbol <rabbit> is symmetrically related to its referent.

Communication is to be regarded as the reliable transfer of information from actor to reactor, resulting in a behavioural change. Symbols are to be seen as one method of information transfer. There are at least four methods of such transfer, shown below in Table 1:

Table 1: Summary of Information Categories

It is worth noting that signals and symbols are to some extent similar. Signals, such as the vervet monkeys’ repertoire of alarm calls, indicate specific objects, events or states of affairs. In the vervet case signals indicate specific predators and subsequently stimulate appropriate behavioural response in other vervets (Cheney and Seyfarth, 1985, 1988). Symbols also indicate specific objects, event and states of affairs and trigger behavioural responses. None the less, the relationship between signal and referent is not symmetrical and is based on affective response so nor is it entirely arbitrary.

It is the properties of arbitrary and symmetrical relationship between symbol and referent that have to be explained by any symbol origins story.

I argue that in order to construct a model of the origins of symbols it is instructive to think about the systems and abilities which might have preceded them, and I suggest the following:

1. Classical and operant learning;
2. Some order of signaling system;
3. A bias to categorize at the whole object level.

There is good reason to assume, both a priori and from comparative data, that these abilities have a more ancient provenance than symbols and symbolic behaviour.

Once we have determined what systems might have been in place in ancestral organisms we need to determine the minimum alterations to environment and/or cognition that would allow for the emergence of symbols. Given that most mutations are not good news for an organism it is wise to avoid catastrophic models of symbol origin of the same order that Bickerton (1990, 1996) has proposed for the emergence of syntax (cf. Studdert-Kennedy, 1992). Bickerton argued for a massive mutation that significantly and fortuitously reorganized cortical architecture. Such an argument to explain the origins of symbolic behaviour would also rely on great fortune rather than minor changes and variance being selected for more cautiously (albeit blindly by Natural Selection).

We could possibly use arguments based on the conception of pre-adaptation (or exaptation as Gould and Vrba, 1982 referred to it).

In order to think about putative models we also need to think about the defining characteristics of symbols as they are used in human language. I argue that the following things are crucial:

1. The vocal medium – most languages are spoken (Locke, 1998) which strongly suggests this was the medium in which ancestral linguistic abilities emerged. Written and sign languages are historical inventions that have arisen after spoken languages so, given this, we must produce a possible explanation of how symbols arose within a vocal medium in the first instance.
2. Social cognition – modern Homo sapiens is a social species, just like many other primate species. From comparative evidence we can hypothesize that our hominid ancestors were a social and hierarchically organized species. Any communication will have happened within this social order. Given the manipulative nature of communication we would expect such communication to reflect social complexity.
3. Ostensive behaviour – there is comparative data to suggest that non-human primates engage in ostensive behaviour – either directly pointing with their hands or utilizing the direction of gaze (Gomez, 1998a,b; Leavens and Hopkins, 1998). Also modern humans use gesture to augment spoken language. This proto-referential behaviour is a communication in its own right.

Dunbar’s (1993, 1996) hypothetical scenario for the emergence of a vocal language as a form of allogrooming is used as a model for the selection of increased vocalization, rather than language per se. I argue (i) there is a need to clearly indicate to whom specific vocalizations are directed; (ii) it is likely that as vocalizations became predominant they were initially coupled with specific aspects of the physical grooming interaction; and (iii) this coupling was achieved through expectancy learning. Thus the hypothesis can be stated as follows:

Ostensive systems, such as pointing and gaze, might have acted to direct the attention of hominid ancestors to the object of vocalizations. First this might have occurred solely within the environment of vocal grooming, as postulated by Dunbar, but once a flexible vocal system was capable of being directed to specific grooming interactions object discriminations would easily follow. The main impact of social grooming would have been vocal control, the main impact of ostension would be that of guiding other behaviours and rendering them communicative. Once firm associations between vocalization and object, event or state of affairs had been established through ostension the specific ostensive act would be unnecessary. This would liberate the vocalization, which would, to all intents and purposes, be arbitrarily linked to its referent. Other symbol properties of displaced reference and symmetry would necessarily follow.

This hypothesis leaves us with a number of options for the implementation of symmetry. Symmetry might come for free. Equally, it might be the result of a two-way learning procedure (i.e. learning <rabbit> to rabbit and vice versa; Place, 1995/6). Alternatively, a variant of Hurford’s Saussurean Sign argument could be made (Hurford, 1989; cf. Oliphant, 1996, 1997).

This hypothesis suggests some specific constraints acting to guide learning toward symbol acquisition. The process so far described represents an historical progression and therefore leaves one remaining question – how did symbol acquisition become systematized in modern Homo sapiens?

I argue that the ability to use these directing constraints was selected for in a Baldwinian fashion thereby instantiating in the modern human infant the ability to rapidly acquire symbols. Such an argument fits well with the work of Bloom and Markson (1998; Markson and Bloom, 1997) who claim there is no dedicated word learning mechanism in the human infant but instead a canalized predisposition to learn words. Canalization implies the directing of more general (but not necessarily totally general) systems toward specific goals thereby obviating the need for domain specific computations.

 Conference site: http://www.infres.enst.fr/confs/evolang/