Modelling language dynamics in Carinthia, Austria K. Prochazka* and G. Vogl University of Vienna, Faculty of Physics, Boltzmanngasse 5, 1090 Vienna, Austria *[email protected]

map © OpenStreetMap www.openstreetmap.org/copyright

Background Use of the minority language Slovenian in Carinthia, One way of monitoring this language shi on a large Most of these simulations use di erential equations Austria has been steadily declining over the past scale is using methods from the natural sciences and look only at the proportion of speakers in the century. Despite supportive measures, language where dealing with big sets of data is common. Lan- whole population with no spatial dynamics. We shi (speakers abandoning use of one language for guage shi has previously been described by com- present a di erent approach: a microscopic mo del another) is taking place, reducing cultural and intel- puter simulations and equations based on models to follow language shi over time and space in lectual diversity. of di usion from solid state physics.[1–3] Carinthia based on data from the Austrian census.

[1] C. Schulze, D. Stau er, S. Wichmann. Commun Comput Phys 3, 271–294 (2008). [2] A. Kandler. Hum Biol 81, 181–210 (2009). [3] N. Isern, J. Fort. J R Soc Interface 11 (2010).

The data from the Austrian census Challenges The Austrian census was the primary method for collecting data on language use in Aus- Creating new data is impossible, thus we need to get the most out of the tria between 1880 and 2001. Data is available historical data available. every ten years between 1880 and 1910 and A page from the 1890 census with 1951 to 2001, plus for some years in between information about the “vernacular the two periods (with varying data quality). language” (Umgangssprache) in each The sample available is very large and detailed, but due to the anonymi- village (Slovenian or German). ty of the census no additional information about individual persons is available.

Ideally, the model should only use parameters which (a) make sense linguistically and (b) can be empirically measured. Testing the model with empirical data is only possible if parameters are used which corre- spond to the data.

Events influencing language use which are not language shi as such (e.g. wars) but visible in the data cannot be captured by the model.

Our approach: A microscopic model

15 67 + = 112 13 8 42

Carinthia is divided into grid cells, Each village ( ) is assigned to a grid cell based on its geographic size 1 × 1 km each. coordinates and speaker numbers are attributed to the grid cells.

To get the number of speakers of a language α in the next year, each grid cell ( ) is updated according to the mathematical rule: Some results: Carinthia 1880–1910 Graphs show absolute change nα(r, t)+Fα(r, t) nα(r, t+1) = ntotal(r, t+1) census data in Slovenian speaker numbers ·nS(r, t)+FS(r, t)+nG(r, t)+FG(r, t) (red = more, blue = fewer). what this means: the number of speakers in the next year  is pro portional to:  the number of current speakers and

 the interaction with speakers in other places Klagenfurt/Celovec Simulated and real data agree Villach/Beljak well qualitatively.

1 km² All model parameters can be numbers of speakers interaction with speakers numbers of speakers    simulation 0 10 20 km calculated directly from the this year in other places in the next year census data, ensuring that the

b model is applicable even in 15 67 situations where data on other factors influencing language 13 112 8 15 use (e. g. perceived status of a + language) is not available or 42 even possible to obtain.

Acknowledgments We thank A. Gehart & W. Zöllner of the Statistik Austria This work is funded by a uni:docs fellowship from the library for providing historical data. University of Vienna.

1