Conceptual

Temporal-Reasoning Knowledge Distillation for Timeline Summarisation

A method that transfers temporal-reasoning ability into a timeline-summarisation model through knowledge distillation: a teacher language model is fine-tuned on temporal-relation tasks and its knowledge is distilled into a student model trained simultaneously to summarise chronological event sequences. It is introduced alongside NarrativeReason, a dataset of temporal relations among whole sequences of events within narratives rather than isolated event pairs, and improves summarisation of long social-media threads where events lack explicit timestamps.