Applied Computational Intelligence Research Unit
Dept. CIS, University of Paisley, PA1 2BE, UK
With the application of evolutionary theory to cultural evolution, it has become somewhat more common to view language change as the result of evolutionary processes on languages themselves. Such a viewpoint typically ignores the role played by individual and social motivations in creating or progressing language changes, but does not necessarily deny that such motivations have some influence (Lass 1997).
We ask if it is possible that the cultural evolution of languages can be adaptively neutral – that language change does not necessarily result in functionally better languages or provide adaptive benefit to the communities using them. We argue against theories that hold that such benefits are required for linguistic diversity and dialects to emerge.
Opposing views are held by, amongst others, Lass (1997) and Nettle (1999). Lass argues against functional explanations of language change at some length. In contrast Nettle argues that, without any social benefits, the neutral evolution of language would be unable to produce significant linguistic diversity or language change. Nettle also builds a computational model to support his case.
We review Nettle’s argument and model, highlighting weaknesses of both, and counter with further arguments to support the neutral evolution of languages. We show that significant and sustained change and diversity can exist without apparent means of selection. Like Nettle, we support our argument with a computational model. In the model, diversity emerges and is maintained over time, while the abstract ‘languages’ continually change and evolve – despite the model lacking any means of selecting for diversity. A brief review is also given of other models which support the notion of adaptively neutral linguistic evolution.
Finally, we consider whether neutral evolution excludes other explanations or causes of language change, and discuss the relationship between these different causes.
Both Lass and Nettle draw on the work of Kimura (1983), who showed that it was possible for evolution to occur without any apparent selective forces at work. However, they give very different explanations of why the neutral-evolution of language should, or should not, be sufficient to cause change and diversity in human languages.
Lass’ proposal is based on the observation that languages are imperfectly replicating systems, within which elements of linguistic ‘junk’ and other ‘marginal’ features exist. This provides ample room for variation, and allows changes to occur without disrupting the success of communication. That replication is not, and can not be, perfect means that languages will change, regardless of functional benefits.
Nettle argues against the neutral-evolution of linguistic systems on three points:
1. Random changes would be non-directional and could be expected to cancel each other out, due to an averaging effect.
2. With a neutral model it is difficult to account for diversification without geographical isolation.
3. Structural correlations in many of the world’s languages represents parallel evolution, showing that the path of linguistic diversification is not random.
Thus, Nettle proposes that in order for linguistic evolution to occur without geographical isolation additional mechanisms are required. Nettle argues that the social functions of language are required for the emergence of linguistic diversity, a view shared by Dunbar:
"… dialects arose as an attempt to control the depredations of those who would exploit peoples natural cooperativeness", Dunbar (1996, page 169).
Nettle’s third point does not relate to this contention, and is outwith the scope of this current work. It does not represent an unsurpassable problem, however. For example, linguistic diffusion seems to indicate that significant randomness exists in language change, despite the apparent regularity of the results. Constraints, innate or otherwise, also impact upon linguistic evolution and may account for many structural correlations (Kirby 1999).
Nettle’s first and second points both rely on the equal distribution of individuals, with a uniform likelihood of any one individual interacting with any other. As recognised by Cavalli-Sforza and Feldman (1978), in any group the amount of influence exerted on any one individual by any one of the others will vary according to a number of factors. This reduces the effect of averaging, and increases the potential for sub-populations to vary from the mean. The different social networks within groups reduces the need for geographical isolation to produce linguistic diversity. The importance of social networks as an influence on language change is emphasised by Milroy and Milroy (1993).
Further, the averaging effect itself is questionable. For example, for random variation in the formant frequencies of phonemes it may not be the case that such variation will ‘cancel out’, or that the average values will be learned. Phonemic and articulatory constraints (see Lindblom, 1998, and de Boer, 1999) may prevent the ‘cancelling out’, and the learned phonemes may be only tolerably close to those heard. With discrete forms, given two different forms of a linguistic feature a learner does not choose just one to learn, but learns both. One may be preferred, but both may be used in varying amounts. As well as applying to the lexicon, it has been proposed that language learners learn multiple grammars, so as to cope with the variation in grammars in use around them (Kroch, 1989).
We now review and criticise the computational model built by Nettle. Then we describe our model, and review others which add strength to the arguments for the neutral evolution of linguistic diversity.
First, in an earlier paper, Nettle and Dunbar (1997) built a model which demonstrates that language diversity can operate as a linguistic marker that serves an important social function. In the case of the model, by excluding non-cooperative individuals from participating in social exchanges. Nettle concludes that its use as a social marker explains the reason why linguistic diversity exists, but this model is merely a demonstration that linguistic diversity can serve a social function. That the utility of dialect diversity is what leads to its emergence is not proven.
The model in Nettle (1999, Chapter 3) attempts to demonstrate how neutral evolution of language cannot lead to diversity, using a computational model in which learners learn phonemes from adults in the same group. There are, however, some questionable characteristics of the model. Language learners in this model learn the average formant frequency values used by the adults in the population, with slight perturbation due to noise. All children in a group learn simultaneously from a single ‘snapshot’ of the adult phonological systems. This learning process explicitly limits the possible variation within a group, and the phonetic systems evolve very slowly.
The model tries to show that inter-group diversity cannot emerge through random change alone. A number of ‘family’ groups exist in the model, and where there is no interaction between groups, inter-group diversity emerges. When individuals migrate between groups, given the explicit averaging and the exceedingly slow rate of change, it is not too surprising that the migration destroys inter-group diversity. Nettle proceeds to show that if learners only learn from selected role-models – a form of social selection – inter-group diversity will again emerge. We shall see, however, that such selection is not required to demonstrate the evolution of linguistic diversity.
In an attempt to show the effect that the geographical or social organisation of language users can have on linguistic diversity, we previously developed a model in which the language learners are spatially distributed along a line (Livingstone and Fyfe 1999). The artificial neural network based agents are implemented similarly to those in Livingstone and Fyfe (2000), but are homogenous in structure. Learners are taught only by individuals in the parent generation within a local neighbourhood.
From random initial conditions, the resultant languages are such that considerable diversity exists globally, but local clusters of agents learn similar languages with low diversity. A dialect continuum is observed in the model – all individuals are capable of successful communication with others in the near vicinity. The likelihood of successful communication decreases with the distance between any pair of communicating of agents. As new generations learn, there is continued fluctuation and change in the learned languages. Thus, without geographical isolation, or any means of functional or social selection, diversity and change are still observed.
One weakness of this model is that starting from non-random initial conditions, where all agents in the first generation have the same language, no changes to the linguistic system take place and diversity does not emerge. However, by adding a small amount of noise to the model it is seen that diversity does emerge and the results are soon indistinguishable from those of random starting conditions. We use noise to represent errors in linguistic transmission and random variations in an individuals own use of language, factors where it is difficult to determine a realistic figure, but believe the value used (below 1%) to be reasonable, perhaps even conservative.
Other computational models, investigating language change in spatially homogenous populations, have shown similar results. In Steels and Kaplan (1998) random variation, caused by stochastic errors introduced into language games between pairs of agents, is able to create competition between linguistic forms in the agent languages, resulting in change over time. Stoness and Dircks (1999) present a similar model, one which does not rely on random noise to maintain competition between forms.
The assertion that some social function is necessary for linguistic diversity and language change to occur is not supported by these models. Change and diversity is caused simply by the (imperfect) replication of language over many interactions between individuals.
We have argued that no functional or adaptive benefits are required to create linguistic diversity, and that diversity should arise naturally from the imperfect transmission of language from users to learners. This represents a neutral theory of linguistic evolution.
We have reviewed some of the objections to a neutral theory, and shown them to be inconclusive. Accordingly, social or linguistic functions are seen to be unnecessary for the emergence of diversity. What then is the role of social and personal motivation in language change? To say that adaptive benefits are not required for the evolution of diversity is not to say that such benefits do not exist, or that they do not influence the evolution of languages. Indeed, classic studies such as that of language change in Martha’s Vineyard (Labov, 1972) show that social factors do exert a strong influence.
Accepting that language changes are influenced by social pressures on language users, we can question why language users adapt their language according to such pressures. Is there something remarkable in the human ability to determine significant social information simply from accent and dialect, without regard to the content of the speech?
Rather than claim that it is the usefulness of such abilities that led to the evolution of linguistic diversity, we argue that the reverse is more likely – that the existing linguistic diversity may have led to the development of such abilities. Humans have had many millennia to adapt to living in societies with linguistic diversity present. That people are able to modify language according to situation means that linguistic change is unlikely to be truly neutral, as social pressures influence which changes will succeed in a population. This does not negate the neutral evolution of language, it adds to it.
We conclude that the neutral evolution of languages is unavoidable and remains a factor in language change, but is not the only cause of change. Yet neither social nor linguistic function are required to create linguistic diversity – geographical spread and imperfect transmission alone are sufficient to account for this.
de Boer, B. (2000). Emergence of sound systems through self-organisation. The Evolutionary Emergence of Language. C. Knight, M. Studdert-Kennedy and J. R. Hurford, Eds., Cambridge University Press.
Cavalli-Sforza, L. L. and M. W. Feldman (1978). Towards a Theory of Cultural Evolution. Interdisciplinary Review 3: 99-107.
Dunbar, R. (1996). Grooming, gossip and the evolution of language, Faber & Faber.
Kirby, S. (1999). Function, Selection, and Innateness: The emergence of language universals. Oxford University Press.
Kroch, A. (1989). Reflexes of grammar in patterns of language change. Journal of Language Variation and Change 1: 199-244.
Labov, W. (1972). Sociolinguistic Patterns. Philadelphia, University of Pennsylvania Press.
Lass, R. (1997). Historical Linguistics and Language Change. Cambridge University Press.
Lindblom, B. (1998). Systemic constraints and adaptive change in the formation of sound structure. Approaches to the Evolution of Language. J. R. Hurford, M. Studdert-Kennedy and C. Knight, Eds., Cambridge University Press: 242-264.
Livingstone, D. and C. Fyfe (2000). Modelling Language-Physiology Coevolution. The Evolutionary Emergence of Language. C. Knight, M. Studdert-Kennedy and J. R. Hurford, Eds., Cambridge University Press.
Livingstone, D. and C. Fyfe (1999). Dialect in Learned Communication. AISB '99 Symposium on Imitation in Animals and Artefacts, Edinburgh, The Society for the Study of Artificial Intelligence and Simulation of Behaviour.
Milroy, J. and L. Milroy (1993). Mechanisms of change in urban dialects: the role of class, social network and gender. International Journal of Applied Linguistics 3(1).
Nettle, D. and R. I. M. Dunbar (1997). Social Markers and the Evolution of Reciprocal Exchange. Current Anthropology 38(1): 93-99.
Nettle, D. (1999). Linguistic Diversity, Oxford University Press.
Steels, L. and F. Kaplan (1998). Stochasticity as a Source of Innovation in Language Games. Artificial Life VI, C. Adami, R. Belew, H. Kitano, C. Taylor., Eds., MIT Press.
Stoness, S. and C. Dircks, (1999) Investigating Language Change: A Multi-Agent Neural-Network Based Simulation, Hahn, M. and Stoness, S., Eds., Proceedings of the Twenty-First Annual Meeting of the Cognitive Science Society, Vancouver, BC.
Conference site: http://www.infres.enst.fr/confs/evolang/