Quantifying the Driving Factors for Language Shift in a Bilingual Region
Total Page：16
File Type：pdf, Size：1020Kb
Quantifying the driving factors for language shift in a bilingual region Katharina Prochazkaa,1 and Gero Vogla aDynamics of Condensed Systems, Faculty of Physics, University of Vienna, 1090 Vienna, Austria Edited by Barbara H. Partee, University of Massachusetts at Amherst, Amherst, MA, and approved February 13, 2017 (received for review November 2, 2016) Many of the world’s around 6,000 languages are in danger of mountain range, the Karawanks, from the neighbor country Slovenia disappearing as people give up use of a minority language in favor where Slovenian is the national language. In southern Carinthia, of the majority language in a process called language shift. Lan- which comprises the districts Klagenfurt and Völkermarkt and parts guage shift can be monitored on a large scale through the use of of the districts Hermagor and Villach (Fig. 1A), the population mathematical models by way of differential equations, for exam- spoke and speaks partly German and partly Slovenian, the territo- ple, reaction–diffusion equations. Here, we use a different ap- ries being intermixed (11). However, the number of Slovenian proach: we propose a model for language dynamics based on speakers in Carinthia has drastically decreased between 1880 and the principles of cellular automata/agent-based modeling and 2001 (Fig. 1 B and C), and language shift is taking place. We use the combine it with very detailed empirical data. Our model makes data from this case to evaluate our proposed model and its as- it possible to follow language dynamics over space and time, sumptions. Checking against empirical data also allows us to ex- whereas existing models based on differential equations average plicitly identify the factors influencing language shift and quantify over space and consequently provide no information on local their impact. changes in language use. Additionally, cellular automata models can be used even in cases where models based on differential Limits of the Classic Macroscopic Reaction–Diffusion equations are not applicable, for example, in situations where Approach one language has become dispersed and retreated to language In the past, language spread and retreat were mostly investigated islands. Using data from a bilingual region in Austria, we show on a macroscale using differential equations. Macroscopic ap- that the most important factor in determining the spread and re- proaches gained popularity after Abrams and Strogatz (12) SOCIAL SCIENCES treat of a language is the interaction with speakers of the same published a short seminal paper in 2003 describing the retreat of language. External factors like bilingual schools or parish language languages with what they called lower status by differential equations. have only a minor influence. Their differential equation system considered only temporal and no spatial development, but the paper has drawn a tail of publications in language shift | diffusion | language dynamics | quantitative linguistics | cellular automata its wake, many of them including spatial development. Spatial and temporal development of languages are usually combined in re- action–diffusion equations (13–19) of the form ∂u/∂t = D·∂2u/∂x2 + ’ COMPUTER SCIENCES t is estimated that around 90% of the world s 6,000 languages f(u). These types of equation are also used in other fields, for Iwill be replaced by a few dominant languages by the end of the example, biology or chemistry, to describe all kinds of spread 21st century (1). This replacement, which is called “language phenomena (20). shift” (2), leads to a loss of cultural diversity. To prevent this loss Considering a language with higher status, for example, German and preserve endangered languages, researchers have been try- vs. Slovenian in Carinthia, the development of the fraction u (x,t) ing to find and quantify the factors behind language shift. Lan- G of German speakers in the total population can be written as a guage shift (speakers giving up use of one language in favor of another) is driven by a variety of influences, for instance, de- mographic and social factors (3–5). To quantify the influence of Significance each of these factors and to study language shift on a large scale, mathematical models and computer simulations have been pro- Languages are an important part of our culturally diverse posed (6, 7). These models generally fall into two categories: world, yet many of today’s languages are in danger of dying (i) macroscopic reaction–diffusion equations that describe the con- out. To save endangered languages, one must first understand centration (fraction) of speakers in the population; (ii) microscopic the dynamics behind language shift: what are the driving fac- agent-based models that simulate the actions of individual speakers tors of people giving up one language for another? Here, we (“agents”) changing their language with a certain probability at each model language dynamics in time and space starting from interaction. For evaluating both types of model, parameters are empirical data. We show that it is the interaction with speakers required that can be empirically measured so that they can be fitted of the same language that fundamentally determines spread to data (8). This means that data covering language use over time and retreat of a language. This means that a minimum-sized and space are needed, but such data are often not available in neighborhood of speakers interacting with each other is es- sufficient resolution. Therefore, mathematical models have so far sential to preserve the language. only rarely been checked against data on actual language use. In this work, we combine mathematical modeling with very de- Author contributions: K.P. and G.V. designed research, performed research, analyzed data, and wrote the paper. tailed empirical data. Applying diffusion theory from physics, we The authors declare no conflict of interest. propose a simple model to describe the dynamics of language shift on a microscopic scale based on the principles of cellular automata/ This article is a PNAS Direct Submission. agent-based modeling (9, 10). The historical data come from Data deposition: The digitized census data on language use for southern Carinthia, 1880– 1910, from the Austrian/Austro-Hungarian census reported in this paper have been de- southern Carinthia, Austria, which provides an extremely well- posited in figshare (https://figshare.com/articles/Language_use_in_Carinthia/4535399). documented linguistic ecosystem with the interaction of two lan- 1To whom correspondence should be addressed. Email: katharina.prochazka@ guages on one and the same territory. Carinthia was a federal state of univie.ac.at. the Austro-Hungarian Empire until 1918 and of the Federal Re- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. public of Austria afterward. It is geographically separated by a high 1073/pnas.1617252114/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1617252114 PNAS | April 25, 2017 | vol. 114 | no. 17 | 4365–4369 Downloaded by guest on September 29, 2021 AB C Fig. 1. Percentage of Slovenian speakers in southern Carinthia according to census results. (A) Geographic location of southern Carinthia. The study area is indicated by the orange rectangle in the map at Bottom Left.(B) Census results for 1880. (C) Census results for 2001. Cells without any Slovenian speakers (0%) are shown in white. Each square marks a 1 × 1-km cell. Contour lines are in brown, and rivers and lakes are in gray. The two biggest Carinthian towns, Villach and the capital Klagenfurt, are encircled according to present administration. reaction–diffusion equation following Fisher’sequationfor prevented consistent censuses and data with the same level of detail is not advantageous genes (21): available. In the censuses together with several other items, the vernacular language of each person was asked for and registered. In the censuses until ∂u =∂t = D · ∂2u ∂x2 + k · u · ð1 − u Þ. [1] 1910, only one language could be recorded in the questionnaire, whereas in G G G G G the censuses after 1971 also bilingualism could be indicated. For the sake of simplicity and consistency and data with the same level of detail is not Here, DG is the diffusivity of German language and k is the con- available, we included bilingual Slovenian–German speakers in the minority version rate from Slovenian to German. The fraction of Slovenian group, that is, Slovenian speakers. In this paper, we do not consider bi- speakers is given by uS = 1 − uG. lingualism as a separate speaker state. We neither consider the different In Carinthia, the language front between German and Slovenian Slovenian and German dialects that are not encoded in the census data. essentially advances only southward motivating a one-dimensional From the census results, we read the number of German and Slovenian treatment. Eq. 1 then results in a traveling front of the higher status speakers in each of the ∼1,500 population units (hamlets, villages, and language (in this case German) with velocity v: towns) in southern Carinthia. pffiffiffiffiffiffiffiffiffiffiffiffi v = 2 D · k. [2] Mapping the Data. For our study of language dynamics in Carinthia, the area G investigated is subdivided into a quadratic grid with cells sized 1 × 1 km. Carinthia spans ∼2.4° of longitude or 184 km (east border to west border) and 0.8° of We defined the front as the line bordering all cells with more latitude or 84 km (north border to south border). We thus cover the total area of than 50% Slovenian speakers each without including outlier cells Carinthia with a regular grid of 84 × 184 = 15,456 cells of 1 × 1 km each. Of these detached from the contiguous language area (Fig. 2A). 15,456 cells, 9,549 comprise the actual area of Carinthia. A total of 1,170 of the From the data, the velocity of the language front can be derived cells inside the Carinthian borders is populated by data for speakers in southern (Table S1) and the product of diffusivity and conversion rate DG·k Carinthia.