Skip to content
conftrace_
2026 AACL AACL 2026

Corrective Retrieval Augmented Generation

AbstractLarge language models (LLMs) inevitably exhibit hallucinations since the accuracy of generated texts cannot be secured solely by the parametric knowledge they encapsulate. Although retrieval-augmented generation (RAG) is a practicable complement to LLMs, it relies heavily on the relevance of retrieved documents, raising concerns about how the model behaves if retrieval goes wrong. To this end, we propose Corrective Retrieval Augmented Generation (CRAG), an autonomous agent that dynamically corrects and refines the retrieval content. Specifically, an agent core is introduced that perceives, decides, and executes in a closed loop. It dynamically selects and orchestrates tools, including internal retrieval, web search, keyword evaluators, semantic evaluators, and LLM evaluators, to perform multi-dimensional contextual quality evaluation and knowledge supplementation, thereby constructing accurate and precise retrieval content. Furthermore, because CRAG refines and corrects retrieved context without requiring modifications to the underlying system, it naturally serves as a plug-and-play framework that can be seamlessly integrated with a wide range of RAG-based approaches. Experiments on five datasets demonstrate that CRAG outperforms previous approaches, while its autonomous decision-making capabilities improve tool-calling efficiency and retrieval context quality.

🌉 Interdisciplinary Bridge - Artificial Intelligence and Natural Language Processing

More from AACL 2026

All AACL 2026 papers →