How Well Do Large Language Models Disambiguate Swedish Words?
This work addresses the problem of word sense disambiguation in Swedish for NLP applications, but it is incremental as it applies existing models to a new language and benchmarks.
The study evaluated recent large language models on Swedish word sense disambiguation benchmarks, finding that while they underperform supervised methods when training data is available, most models surpass graph-based unsupervised systems, with the best accuracy achieved by including human-written sense definitions in prompts.
We evaluate a battery of recent large language models on two benchmarks for word sense disambiguation in Swedish. At present, all current models are less accurate than the best supervised disambiguators in cases where a training set is available, but most models outperform graph-based unsupervised systems. Different prompting approaches are compared, with a focus on how to express the set of possible senses in a given context. The best accuracies are achieved when human-written definitions of the senses are included in the prompts.