Table 3.

Examples of comparison of (1) labeled (ground-truth) sections related to symptoms and (2) sections estimated by large language models within given transcript segments. The estimation becomes more accurate (ie, there is a greater overlap between the ground-truth and estimated sections) as the corresponding mid-token distance (d) decreases.

Recall mid-token distance d	Ground-truth section	Estimated section
0	But when I dream about it, I dream about the scene of my escape, the scene of my escape from North Korea, the scene of my escape from the police, and I still dream about it.	But when I dream about it, I dream about the scene of my escape, the scene of my escape from North Korea, the scene of my escape from the police, and I still dream about it.
2	Memory I don't really want to think about	I don't know, I haven't pulled it out in a long time, and it's actually a memory I don't really want to think about. Yeah.
6	Yes. That's hard and scary too.	It sounds like it's hard for you to be intimate with guys and have new relationships and stuff like that. P3: Yeah. That's hard and scary too.
31	That's what I still think about now, why did I say that, when he's gone, why did I say that, and that's what I regret.	Never the things of my heart. I am unjust. My heart is broken. I'm hurting. I'm just not expressing it.
Infinity	It's because we're conditioned to think that anyone in black is someone who's out to get us.	None