Asymmetric Calibration of Social Support

A taxonomy of support-seeking and support-giving moves, applied to a corpus of nearly 8,000 Reddit dyads. We show that the two sides of a support conversation calibrate asymmetrically, with implications for how LLMs should generate and evaluate empathetic responses.

Full project write-up coming soon.