SILD
We have successfully acquired funding for a four-year DFG Research Unit ("Semantic Shifts in Low-Resource Domains" – SILD / FOR 5862), co-chaired by Prof. Dr. Andreas Hotho and Prof. Dr. Fotis Jannidis (Chair for Computer Philology and History of Contemporary German Literature). The goal of the project is to improve the capabilities of Language Models to detect and explain changes in word meaning across historical and specialized texts, an area where current models often struggle due to limited training data and a lack of interpretability.
Project Overview
SILD is an interdisciplinary collaboration between computer science, computational linguistics, visual analytics, and the humanities (including collaborators from the Universities of Würzburg, Erlangen-Nuremberg, and Konstanz). In the humanities, tracking how the meaning of words and expressions changes over time is a fundamental task, whether in early modern German drama, Medieval Latin translations, or regional varieties like Hong Kong English.
Within our subproject HELM, we focus on combining Language Models with structured linguistic knowledge from domain-specific resources such as dictionaries and encyclopedias. We aim to semi-automatically construct domain-specific Knowledge Graphs and tightly integrate them with Language Models to improve language representations in settings where only small amounts of text are available.
The project will significantly advance the state of the art in the combination of Language Models and Knowledge Graphs, showing how to model and analyze semantic shift by (a) integrating expert domain knowledge into Language Models and (b) automatically updating Knowledge Graphs based on text analysis to suggest new candidates for meaning change. Through an iterative feedback loop with humanities scholars, we aim to make semantic shifts between language domains directly visible, verifiable, and reusable for both specialized and general-purpose Large Language Models.
