In a recent study (Baumann et al. 2023, EMNLP), we found, based on English diachronic corpus data, that concrete words are more likely to become polysemous than abstract words; a plausible mechanism for this is that concrete words are more likely to yield metaphorical extensions than abstract words. The goal of this project is to test if the same relationship holds in German. For this, contexts of words will be extracted from DWDS (https://www.dwds.de/d/korpora/kern) for all decades in the 19th and 20th century, and subsequently analyzed via BERT-based sentence embeddings to derive a measure of increase/decrease in polysemy. Finally, this measure will be correlated with lexical concreteness ratings (Charbonnier & Wartena 2020, SWISSTEXT).

Prerequisites: R or Python, text processing, ideally some familiarity with BERT, statistical analysis 

Languages are constantly subject to change. This is most clearly visible in the words that we use. While some words enter the lexicon (i.e., the shared vocabulary of a speech community) and become more frequent, others get less frequent and might even vanish. But do these processes apply uniformly across languages and do words spread faster in languages with more strongly connected speech communities? The goal of this project is to examine rates of lexical change in different languages. For this, a size-balanced diachronic multilingual corpus is first constructed by sampling texts from Wikipedia. In a second step, rates of lexical change and fluctuation are measured through statistical analysis in all languages. Finally, measures of how strongly speakers are connected to each other (e.g., population density) will be correlated with the average rates of lexical change per language.

Prerequisites: R or Python, ideally experience with text processing and the Wikipedia API, statistical analysis

This project is part of  the DIGILINGDIV project. Students are expected to participate in project meetings on a regular basis.

Compartmental ODE models established in the field of epidemiology have been used to study the diffusion of linguistic innovations, such as the spread of new words. Most approaches employ models in continuous time (Nowak 2000, J. Math. Biol., and research of my own). The goal of this project is to examine to what extent compartment models in discrete time (i.e., systems of difference equations) can be reasonably transferred from epidemiology to the study of language change. One interesting feature of some discrete-time models studied in epidemiology (in particular, SIS models) is that they can display periodic behavior, i.e., periodically fluctuating trajectories (Allen 1994, Mathematical Biosciences). Do we find such behavior in linguistic dynamics and can discrete-time models be meaningfully employed to account for it?

Prerequisites: mathematics of dynamical systems; familiarity with R, Python, Mathematica, Matlab, or similar; maybe a bit of diachronic corpus analysis and/or time series analysis

This project was about a dataset consisting of acoustic features of vowels (first and second formant) that have been produced by white and colored speakers of Afrikaans, a Germanic language that is spoken (mainly) in South Africa. The data have been recorded using two different methods (word lists vs. speech in context) by Daan Wissing (North West University, Potchefstroom, South Africa). Acoustic features have been already extracted for all configurations. The goal of the project was to compare the different configurations to assess (a) whether the examined sociolinguistic variants of Afrikaans differ from each other and (b) to what extent results based on different methodologies match.

Global linguistic diversity has declined, but the mechanisms behind this trend remain unclear. While past research focused on socioeconomic factors, this study examines the role of digital literacy. Limited digital proficiency is hypothesized to marginalize minority languages online. To test this, we compiled global language data and used Shannon entropy to quantify linguistic diversity, analyzing links with digital and demographic indicators. Our findings show that internet access supports linguistic diversity, while education levels negatively impact it. However, digital skills had no significant effect. The study contributes to discussions on language endangerment and suggests directions for future research.