How Well Do Large Language Models Disambiguate Swedish Words?

TL;DR

This paper evaluates large language models' ability to disambiguate Swedish words, comparing different prompting methods and highlighting the importance of human-written sense definitions for accuracy.

Contribution

It provides a comprehensive assessment of recent large language models on Swedish word sense disambiguation benchmarks and explores effective prompting strategies.

Findings

Models outperform unsupervised systems but lag behind supervised disambiguators with training data.

Including human-written sense definitions in prompts improves accuracy.

Different prompting approaches significantly impact disambiguation performance.

Abstract

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.