Korean neologisms emerge through a variety of mechanisms, which makes them a demanding test for large language models. Unlike fixed vocabulary, these new words require models to go beyond lexical memorization and handle word formation processes in context.
To support systematic evaluation, the authors of the arXiv preprint introduce NeoTrace, an automated and rerunnable framework that collects and dates neologisms. The framework is designed to enable repeatable assessment rather than one-off benchmarks.
The paper's title poses three questions: do LLMs know Korean neologisms, can they learn them, and can they coin new ones? The abstract available here does not include specific results, so the article cannot report findings beyond the stated research aims.