Dutch Metaphor Extraction from Cancer Patients’ Interviews and Forum Data Using LLMs and Human in the Loop

Metaphors play an important role in communication between cancer patients and clinicians.In this work, we present the first study on automated extraction of Dutch metaphors from cancer patient interviews and online forum data using large language models (LLMs) and human validation.We develop a human-in-the-loop extraction framework combining structured prompting, automatic verification, and expert linguistic evaluation.We evaluate multiple open-source LLMs and prompting strategies, including chain-of-thought and knowledge-guided prompting.Our approach resulted in the HealthQuote.NL dataset, containing 130 validated metaphors.We analyse common model errors, including hallucination and confusion with idioms, and show that refined prompting achieved 63.2% precision.This work provides both a novel dataset and an extraction framework to support patient-centred healthcare communication and future metaphor research.We share our prompts and resources at https://github.com/4dpicture/HealthQuote.