Computerlinguistik-Tutorium: Lexical Semantic Change Detection

Di, 4.3.2025, 13:00-17:00 Uhr (max. 45 Teilnehmende, Anmeldung über Konferenzanmeldeformular erforderlich)

Lexical Semantic Change Detection (LSCD) aims to automate the identification of word sense changes within diachronic text corpora. It has its applications in historical linguistics and lexicography, but also in the social sciences where distinctions in language use across groups of interest (such as political parties) can help to provide evidence for research hypotheses. However, limited programming knowledge and missing standardization can be obstacles in applications of LSCD models. Hence, I will introduce two softwares aimed at simplifying the application of these models: The first one is the DURel tool [1]. DURel is an online user-interface for annotating sentence pairs of words, both with humans and computers. Annotations of words are used to create clusters of meanings that can be visualized over time. This way, lexical properties such as polysemy, vagueness or the change in meaning of words can be examined. The second software is the LSCD Benchmark implementing evaluation procedures for models on most available datasets in Python [2]. With simple commandline inputs the user can specify model, dataset and metric for evaluation. New semantic change measures can easily be added to the benchmark because of its modular implementation. I will introduce the necessary theoretical background and demonstrate the usage of both softwares in the tutorial in order to give participants a starting point to apply them in their research.

References
[1] https://durel.ims.uni-stuttgart.de/
[2] https://github.com/ChangeIsKey/LSCDBenchmark