An Approach for Systematic Decomposition of Complex LLM Tasks

ACL ARR 2026 January Submission7554 Authors

Keywords: large language models, task decomposition, constraint satisfaction, complexity measures, reasoning reliability, combinatorial reasoning, database querying

Abstract: Large Language Models (LLMs) suffer from reliability issues on complex tasks, as existing decomposition methods are heuristic and rely on agent or manual decomposition. This work introduces a novel, systematic decomposition framework that we call CONstraint-Induced Complexity (ACONIC), which models the task as a constraint problem and leverages formal complexity measures to guide decomposition. On combinatorial (SATBench) and LLM database querying tasks (Spider), we find that by decomposing the tasks following the measure of complexity, agent can perform considerably better.

Paper Type: Short

Research Area: LLM Efficiency

Research Area Keywords: parameter-efficient-training, LLM Efficiency, data augmentation

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Approaches low compute settings-efficiency, Data analysis

Languages Studied: English

Submission Number: 7554