A research team is developing a system to analyze long, unstructured patient interview transcripts to identify shifts in emotional state. A key step is to first divide the transcripts into coherent, semantically-related segments. The team is debating between two implementation strategies: 1) using a pre-existing, specialized linguistic segmentation system known for its speed and efficiency, and 2) prompting a large, general-purpose language model to identify and mark the segment boundaries based on the conversational flow.