TypeCoT at UZH Shared Task 2026: Reconstructing Argumentative Structure in UN Resolutions using Type-Informed Chain-of-Thought

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, short

Summary

TypeCoT, a submission to the UZH Shared Task at the ArgMining Workshop 2026, presents a modular pipeline for reconstructing the implicit argumentative structure within United Nations and UNESCO resolutions. This challenging natural language processing task involves analyzing highly structured preambles and operative clauses. Adhering to a strict constraint of using open-weight models with at most 8B parameters, the TypeCoT system is built upon the Qwen-2.5-7B-Instruct architecture. For Subtask 1, it decouples the problem, utilizing a 4-bit quantized LoRA adapter via the Unsloth framework for paragraph type classification, and a type-informed chain-of-thought approach for subsequent thematic tagging and relation prediction.

Key takeaway

For NLP Engineers and AI Scientists tasked with reconstructing complex document structures under strict resource constraints, TypeCoT demonstrates an effective modular approach. You should consider decoupling such problems into distinct classification and reasoning stages. Employing 4-bit quantized LoRA adapters with frameworks like Unsloth can enable efficient fine-tuning on smaller models, while a type-informed chain-of-thought strategy can enhance accuracy for thematic tagging and relation prediction in highly structured texts like UN resolutions.

Key insights

A modular, type-informed chain-of-thought pipeline efficiently reconstructs argumentative structure in UN resolutions using constrained open-weight models.

Principles

• Decouple complex NLP problems.
• Adhere to strict model parameter constraints.
• Utilize type-informed reasoning.

Method

The pipeline uses Qwen-2.5-7B-Instruct. It employs a 4-bit quantized LoRA adapter via Unsloth for paragraph type classification, followed by a type-informed chain-of-thought approach for thematic tagging and relation prediction.

In practice

• Implement 4-bit quantized LoRA for classification.
• Apply chain-of-thought for complex tagging.
• Consider Unsloth for efficient fine-tuning.

Topics

• Argument Mining
• UN Resolutions
• Chain-of-Thought
• Large Language Models
• LoRA
• Model Quantization
• NLP Pipelines

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Revisiting Semantic Role Labeling: Efficient Structured Inference with Dependency-Informed Analysis — A new SRL framework offers explicit predicate-argument structure and 10x faster inference with modern encoders.
• NTS-CoT: Mitigating Hallucinations in LLM-based News Timeline Summarization with Chain-of-Thought Reasoning — NTS-CoT uses Chain-of-Thought reasoning to reduce hallucinations and omissions in LLM-based news timeline summarization.
• XDomainBench: Diagnosing Reasoning Collapse in High-Dimensional Scientific Knowledge Composition — LLMs exhibit systematic reasoning collapse in complex, interactive scientific knowledge composition tasks.
• ADaPT: Token-Level Decoupling for Efficient Large Reasoning Models — ADaPT decouples efficiency and correctness in large reasoning models via a token-level dual-process framework.
• HKJudge: A Legal Discourse-Annotated Corpus for Interpreting What Courts Find, How They Reason, and What They Rule — Expert-annotated legal discourse corpora enable advanced AI understanding of court judgments.
• Construction of Knowledge Graph based on Language Model — Lightweight LLMs can construct Hyper-Relational Knowledge Graphs with performance comparable to GPT3.5.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.