Automating Linguistic Frameworks with LLMs: An Approach to Analyze Personal Narratives

Analyzing personal narratives provides insight into how authors shape their identity and construct their reality. This topic has been extensively studied in stylistics as well as in psycholinguistics, where narrative analysis serves as a valuable tool for psychiatric assessment. Yet, formal approaches that integrate a rigorous linguistic framework with the capabilities of generative AI remain scarce. The current paradigm is largely detection-oriented. In this paper, we propose an alternative perspective: we leverage LLMs to automate the extraction of meaningful features inspired by linguistic frameworks, offering a novel way to interpret personal narratives. Our feature set is grounded in Halliday’s Systemic Functional Linguistics and recent psychiatric studies. We first describe our automation methodology, then focus primarily on evaluating the accuracy of the extractions using an evaluation dataset, while providing some preliminary observations on clause segmentation, which will be addressed in detail in a future work. Finally, we compare two LLMs–representing two use cases of the pipeline–using the proposed evaluation methods.