Language organizes meaning through networks of relations among words. Traditional lexical semantics studies synonymy, antonymy, hierarchies, and fields, while historical linguistics compares Indo-European languages using shared innovations. Recent computational methods model meaning via distributional embeddings and analyze system-level structure with semantic networks. Yet links between semantic structure and genealogical relatedness remain underexplored: prior work often relies on lexical overlap or parallel translations, capturing alignment rather than internal organization. This thesis bridges these strands by deriving semantic networks from word embeddings for Indo-European languages and examining graph-theoretic properties. It asks whether related languages share network characteristics and whether such properties recover recognized subgroupings, using hierarchical clustering over structural profiles for comparison.

The ongoing digital transformation is reshaping people's behavior and communication, creating tension between digital communication and linguistic diversity. While a global decline in linguistic diversity is observed, digital platforms like social media fundamentally alter linguistic interaction conditions, differing from analog situations. This is especially relevant in multilingual contexts where speakers regularly switch between different languages. The key question is how digital spaces influence language choice and use, particularly for minority language speakers whose behavior varies across social, cultural, and functional contexts. Digital media offers dual opportunities: it can increase minority language visibility but often favors dominant languages. This research project will examine language choice in communication contexts and develop a standardized questionnaire to assess language choice and motivation in multilingual settings.

This MA-thesis is part of the DIGILINGDIV project.

Polysemous words carry multiple senses, such as German Maus (rodent or computer device). Most words are polysemous, with senses evolving as language changes. Quantifying senses lacks consensus among lexicographers. Polysemy affects language broadly: concrete words more often become polysemous, and it facilitates language acquisition in children through mental sense overlap and foreign language learning. This thesis aims to identify senses contextually for German and verify these relationships. It extracts target word contexts from the digitale Wörterbuch der deutschen Sprache across all decades of the twentieth century, selecting words based on existing research on concreteness and age of acquisition. Using the BERT-based sentence embeddings and the unsupervised k-means clustering, the study derives the sense counts and the polysemy development trends, finally correlating these with the age of acquisition and the concreteness to analyze the behavior of polysemous words in German.

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). 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?

This thesis explores the connection between video gaming and digital reading skills, focusing on 13-year-old students. As digital literacy becomes essential, gaming—often seen as mere entertainment—is recognized for developing key e-literacy skills such as navigation, problem-solving, and information processing. Previous research suggests that gamers, especially those who play single-player games, may outperform non-gamers in digital reading due to enhanced navigation and comprehension skills. The study aims to compare reading performance and digital navigation between gamers and non-gamers to assess this link.

In literary history, there are few figures who have been received as frequently as classical characters. However, evaluating a character’s reception history often depends on the person doing the Interpretation and can thus be limited by their individual understanding of language. While comparing different interpretations is one way to address this problem, I want to try a different approach: a diachronic Emotion analysis using word embeddings from different time periods along with the VAD (Valence-Arousal-Dominance) emotional model. In this way, the resulting VAD scores should better reflect how a text judged a Character using the language of its own time. By comparing works from different periods but centered around the same group of classical women, I hope togain new insights into their reception history.

Authorship attribution seeks to identify the writer of an anonymous text using various methods, now advanced by computational tools like natural language processing, machine learning, and AI. A growing research focus is emotional attribution—analyzing emotional language to uncover unique writing patterns. By integrating emotional tone as a variable, this approach could enhance traditional methods and improve accuracy in identifying authors. The core research question of this thesis is whether emotional features in texts offer reliable clues about authorship. If successful, this method could provide deeper insights into individual writing styles and expand how researchers understand and differentiate authorial voice in text analysis.

This thesis explores how linguistic diversity can be more accurately quantified by integrating interlinguistic distance measures. While previous studies, such as Bromham et al. (2022, Nature Ecology & Evolution), have predicted major language loss using statistical models, they often overlook digitization and similarity between languages. This thesis addresses these gaps using global linguistic databases like PHOIBLE, Grambank, ASJP, and URIEL+. By combining these with Ethnologue speaker data, the project calculates both naive and similarity-weighted diversity measures. It aims to assess data availability, biases, and correlation across phonemic, syntactic, lexical, and other diversity types.

Linguistic diversity is declining rapidly, with nearly half of the world’s 7,000 languages endangered and predictions suggesting up to 90% may disappear by the end of this century. While socio-political, economic, and environmental drivers of language loss are well-documented, the impact of digitization remains understudied. Digitization, a force behind globalization, exacerbates inequalities and contributes to a “digital language divide,” favoring dominant global languages like English. However, it also offers tools for language preservation through apps, AI, and digital platforms. This thesis explores digitization’s dual role by analyzing how digitization indices correlate with linguistic diversity measures, aiming to uncover nuanced patterns and implications.