Narrative-UFET: Narrative Generation for Ultra-Fine Entity Typing

The authors introduce Narrative-UFET, a controlled extension of ultra-fine entity typing that pairs entity mentions with automatically generated short narratives to address limitations in long-tail type disambiguation. The study demonstrates that narrative context yields consistent improvements over sentence-level baselines, particularly when the entity's type shifts within the text.

• Narrative-UFET pairs entity mentions with coherent synthetic narratives to isolate discourse properties.
• Two variants were tested: Maintain (constant type) and Change (shifting type).
• The Change variant provided a stronger signal for improving long-tail type accuracy.
• Synthetic narratives yielded stronger gains than naturally occurring contexts, revealing implicit signals.

The findings indicate that controlled discourse construction can surface disambiguation signals that are often left implicit in real text, highlighting open directions for discourse modeling and narrative construction.