Potential and limitations of LLMs for augmenting lexical knowledge bases
This paper tests whether large language models can extend lexical knowledge bases. Human evaluators accepted 86.7% of novel concepts. Automatic overlap metrics missed many valid additions.