Virtual Linguistic Societies - a Reconstruction of a Historical Scenario

  1. 1. Christer Johansson

    Department of Linguistics - Lund University

Work text
This plain text was ingested for the purpose of full-text search, not to preserve original formatting or readability. For the most complete copy, refer to the original conference program.

“If, however, we find a particular period especially fertile in linguistic changes (…), it is quite natural that we should turn our attention to the social state of the community at that time in order, if possible, to discover some special favouring circumstances. I am thinking especially of two kinds of condition which may operate. In the first place, the influence of parents, and grown-up people generally, may be less than usual, because of an unusual number of parents may be away from home, as in great wars of long duration, or may have been killed off, as in the great plagues … “([4], XIV.—§5, p.260.)

As Otto Jespersen noted the plague years are an interesting period for research in language change. The question is how could we possibly find out how the changes came about. One means of addressing this question is to build an artificial model of the factors that we think are important for the language change to happen. We have already identified some: learning times of transmission from an older generation to a younger, the age distribution in the population, and social mobility. These are factors that are bound to interact in the formation of the ever changing language.

Individual variation is a factor whose influence we need to estimate. In Jespersen’s time there were no computers on which to conduct large scale simulations of populations of interacting speakers. It is therefore not surprising that he stopped at the general statement that there might be an effect by the disturbance of the transmission pattern.

In a larger scale simulation of the plague years ([5], pp. 111-173.) it was found that the actual change in age distribution was quite temporary although increased fluctuations were found throughout the period. Furthermore, under the condition that the population was isolated from outside influence, former distinctions could actually be strengthened. The more elaborate scenario is that the plagues interacted with the influence of the low-German Hansa along the Scandinavian coast ([5]).

Several factors need consideration in order to build up the model. The easiest factor to model is the population development during this period. This has been done by among others Walløe [12] and Bois [1] and further elaborated in [5]. The more difficult part involves what is transmitted in a language learning situation, how this is done and the maturation times involved.

The starting point is that language is transmitted through the perceived distinctions between linguistic elements. This should be fairly non-controversial since this is the standard procedure for finding various units of linguistic theories. A standard procedure is to contrast (minimal) pairs of candidate strings of speech sound (or morphemes or words) and investigate the change in reference. The unit cannot be transferred as a definition, because we cannot know what it refers to without some measure of advanced mind-reading [9].

The distinction between different linguistic forms can be transmitted by the contrasts that they make with other units, without immediate reference to any referred meaning of those linguistic forms. This separation of levels is argued to be a contributing cause of morphological gaps ([5], pp. 92-96.).

Once a unit has emerged (by contrasts within its domain) it can be loaded with a meaning if its expression is consistently used with contrasting functions (i.e., functions similarly exist in contrast to other functions). The function of an expression can thus be viewed as primarily unconnected to expression, but through association between independently formed domains become connected in linguistic signs. Things and acts are not what we call them, and yet we come to discover relations between linguistic markers and possible states of affair to such an extent that we have difficulty keeping them apart.

Physical continuities can be handled as discrete units. For example, we might perceive certain sequences of sound frequencies as instances of different categories. We are not the only creatures that do this [7]. A recent debate has focused on the nature of overgeneralisations by children [8].

Let us consider some ways of learning from a sample. First we could just copy the sample and that would be it. Such a mechanism would not have much individual continuity in the presence of variation. The experience of the individual must somehow be weighted into the learning. This could be done in a blending fashion by taking weighted averages between previous experience and the sample. Such a simple system would be linear, which means that any linguistic innovation would gradually disappear in a population in the blur of mixed distinctions. Overgeneralisation added to the process will make distinctions protected, by enhancing strong distinctions and reducing weak distinctions. An enhancement in perception of distinct forms will mean an enhanced reproduction of that distinction, and vice versa for reduced distinctions. Language is discovered, and then reproduced, by its speakers.

A linguistic system can be understood to be a weakly particulate system. The particles of language are dependent on the variation of the speakers of the language. Consider two sugar cubes which are separate units at normal temperatures, but blend with increased temperature. Temperature is analogous to increased variation in the language population.

Following this reasoning, the model for the transfer between individuals should involve (1) the distinction of an individual, (2) a sample of the distinction from part of the population, (3) overgeneralisation of the sample, and (4) age dependent blending of this with the previous version of the individual’s distinction. Due to overgeneralisation of the sample, and the conformity structure of a society of communicating individuals, the linguistic units are insensitive to small changes in ‘temperature‘, and behave as particles under normal conditions. Blending and particle formation by overgeneralisation explain why there is very little variation in the grammatical possibilities in a linguistic community.

Chomsky has pointed out that language is both hard to explain as an adaptive advantage for the individual, and that it deviates from expected variation in grammatical abilities:

“…language is designed as a system that is “beautiful“ but in general unusable. It is designed for elegance, not for use, though with features that enable it to be used sufficiently for the purposes of normal life. […] Thus it has often proven to be a useful guiding intuition in research that if some property of language is “overdetermined“ by proposed principles, then probably the principles are wrong, and some way should be found to reconstruct them so as to avoid this redundancy. […] Typically, biological systems are not like this at all. […] Why language should be so different from other biological systems is a problem, possibly even a mystery.“ ([3], pp. 49-50.)

The maturation time of a morphological system is dependent on how samples are blended with previous experience. If we can determine the maturation times we could also find the blending parameters. The blending is likely age dependent, since children and adults learn new languages with different speeds. Some investigations [10],[6] have helped to set some approximate values for the blending factors ([5], pp. 122-127.). Children, generally soak up language more quickly, i.e., the sample has a larger impact for them. A 10 to 1 sample reliance results in the approximate observed timing. A slower blending was chosen for adults, stabilising the population around a common value.

Further evidence for a partial blending situation is that the grammatical variation within a population of speakers is fairly low, but not totally absent. Connected to this is the fact that languages can change fairly rapidly from morphological to non-morphological means of marking for case. A mixing of languages in a population could lead to something that is neither from language A nor B if distinctions are dissolved.

Language has had its own evolution. The most significant adaptive advantage of language is for language itself. Although the results of computer simulations do not depend on this fact, it can prove interesting to view the artificial society that is built up in the computer as an environment that can support the formation and selection of certain distinctions (e.g., from the variation in speech). This partly explains the regularities and complexities of languages, since all languages have had several thousands of years to crystallise. The population of speakers is thus seen as the selecting environment of language which is an implicit part of how the computer simulations are designed. A convenient metaphor is to view language as a virus dependent on a population that reproduces it. Language is a cultural virus, analogous to biological ones. Differences in languages appear because some groups are more or less isolated, and therefore have preserved slightly different variations. The differences between groups are then easily confused with difference in expression, which leads to the social discrimination we can observe today of already formed dialects and languages ([5], pp. 69-76.).

Given these conditions for the simulations it is possible to explore why German case marking did not succeed Scandinavian case marking. By introducing a separate marker for the distinctness of the German system, and transfering that in parallel, we can have an estimation of how those markings travel. One factor is the larger Scandinavian population. The third or fourth generation Germans would become more integrated into the Scandinavian language community.

(See QuickTime movies at

Simulations are used where analysis is complex. The complexity is possible to investigate by computational reconstruction. The general critique of such reconstructions is that the results depend on the input to the model, but this is a critique that is common to all analytical models. A second critique is that such models are too general, but this could also be a strength. A third critique is that the resolution in detail is weak in a general model.

However, detailed patterns can be studied and intuitively understood by, for example, presenting the results in time using QuickTime movies. Effects that where previously hard to analyse because they were distributed over populations could be analysed by replicating the population structure. The main advantage of reconstructions is that they provide a concrete, structured meta-discourse in which it is possible to discuss validity and relevance of results and assumptions.

The reconstructions are similar to work by Steels [11], although Steels equates languages and their vocabularies. His language task is parallel to Quines [9] ‘gavagai‘ situation. (Quine demonstrates that language learning cannot be at the level of words and sentences.) Steels does not include age dependent learning and the relation between learning and linguistic complexity of the task, as well as geographical constraints. He neither attempts to reconstruct any historic scenario in his simulations. Without reconstructions, the results are general and abstract, and their validity is often restricted to validity of logical proofs within the theory. The mathematical properties of cultural transmission have been more formally described in [2]. Their model gives some useful analytical proofs that are well-known in population biology but maybe not in linguistics.

Simulating effects in a population over space and time makes it possible to follow the time course as well as the geographical spread in much more detail than in any previous model. One result show how it was possible (and likely) for some specific regions to keep an older marking system. Another result shows how Scandinavia did not become a German speaking area, while loosing case marking. Alternative models of the learner could be compared in a simulation model of a fixed scenario with a known outcome. Detailed predictions can be accomplished since the model is based on observable entities such as the rates of immigration, births and deaths, and the observable development of linguistic markings. The performance of individuals is hard to predict, even within the simulations, due to the individual history of each individual.


1. Bois, G. 1984. The Crisis of Feudalism: Economy and Society in Eastern Normandy c. 1300 ~ 1550. Cambridge/Paris: Cambridge University Press/Editions de la Maison des Sciences de l‘Homme.

2. Cavallai-Sforza, L.L. & Feldman, M. W. Cultural Transmission and Evolution: a quantitative approach, monographs in population biology, 16, Princeton, NJ, Princeton University Press.

3. Chomsky, N. 1991. Linguistics and Cognitive Science: Problems and Mysteries. In A. Kasher (Ed.), The Chomskyan Turn (pp. 26-53). Cambridge, Mass.: Blackwell. Gould, S.J., & Lewontin, R.C. 1979. The spandrels of San Marco and the Panglossian paradigm: a critique of the adaptationist programme. Proc. Royal Society of London, 205(B), 581-598.

4. Jespersen, O. 1950. Language — its nature development and origin. London: George Allen & Unwin Ltd.(first published 1922)

5. Johansson, C. 1997, A View from Language: growth of language in individuals and populations, Travaux de l'Institut de Linguistique de Lund, 34, Lund: Lund University Press.

6. Johnson, J.S., Shenkman, K.D., Newport, E.L., & Medin, D.L. 1996. Indeterminacy in the grammar of adult language learners, Journal of Memory and Language, 35, 335-352.

7. Kluender, K.R., Diehl, R.L., & Killeen, P.R. 1987. Japanese quail can learn phonetic categories, Science, Vol. 237 (pp. 1195-1197).

8. Marcus, F.M., Pinker, S., Ullman, M., Hollander, M., Rosen, T.J., & Xu, F. 1992. Overregularization in Language Acquisition, Monographs of the Society for Research in Child Development, Vol. 57:4. Chicago: University of Chicago Press.

9. Quine, W. V. 1960, Word and Object, Cambridge, MA: MIT Press.

10. Slobin, D.I., & Bever, T.G. 1982. Children use canonical sentence schemas: A crosslinguistic study of word order and inflections. Cognition, 12, 229-265.

11. Steels, L. 1997, Language learning and Language Contact, in Daelemans, W. (ed.) proceedings of the workshop on empirical approaches to language acquisition. Prague. (preprint available: publications.html).

12. Walløe, L. 1995. Plague and Population (Ms Alison Coulthard, Trans.). Oslo, Norway: Det Norske Vetenskabs-Akademi / Inst. of Basic Medical Sciences.

If this content appears in violation of your intellectual property rights, or you see errors or omissions, please reach out to Scott B. Weingart to discuss removing or amending the materials.

Conference Info

In review

"Virtual Communities"

Hosted at Debreceni Egyetem (University of Debrecen) (Lajos Kossuth University)

Debrecen, Hungary

July 5, 1998 - July 10, 1998

109 works by 129 authors indexed

Series: ACH/ALLC (10), ACH/ICCH (18), ALLC/EADH (25)

Organizers: ACH, ALLC