Broadly speaking, I use computational approaches to study how language reflects social and cultural change, treating it as a window into the human mind and society. My work develops theory-driven measures to quantify linguistic, psychological, and social constructs by integrating insights from linguistics and psychology with methods from computational linguistics and Natural Language Processing (a subfield of Artificial Intelligence). Because labeled data are scarce in the social sciences, I primarily use pretrained language models, unsupervised learning, normed lexical resources, and statistical modelling. My current focus is on tracing how societally relevant concepts evolve in meaning over time using large language models and historical text corpora.

With my PhD supervisors, I have developed a linguistic framework (SIBling) and associated measures to model lexical semantic change (LSC) along three major dimensions that are typically overlooked by existing approaches.

Key Contributions:

• SIBling: A theoretical model integrating insights from historical linguistics and psychology, reducing six types of LSC to three core dimensions: Sentiment, Intensity, and Breadth (SIB). [Prototype]
• SIB Toolkit: A computational implementation of SIBling that quantifies semantic change across SIB, and complementary features (salience and thematic content). Designed for broad application across the social sciences and language domains (scientific, media, everyday).
• LSC-Eval: An evaluation framework that uses LLM-generated synthetic corpora to simulate kinds of LSC and validate LSC detection methods, identifying optimal dimension- and domain-specific approaches. [Prototype]
• Applications: Applying SIBling to trace the historical semantic evolution of mental health-related concepts (e.g., autism, schizophrenia), my research examines broader cultural dynamics such as concept creep, pathologisation, and stigmatisation.

This program: (1) offers a multidimensional model of semantic change (SIBling), (2) develops computational tools for its application (SIB Toolkit), (3) establishes a principled evaluation framework for LSC detection methods (LSC-Eval), and (4) demonstrates its value through detailed case studies. Together, these efforts lays the groundwork for future extensions across disciplines (e.g., law, humanities), domains, and languages.