Knowledge-Intensive NLP

NLP tasks that fundamentally require access to external knowledge beyond what is captured in a language model's parameters. Common knowledge-intensive tasks include open-domain question answering, fact verification, knowledge-grounded dialogue, and abstractive question answering. These tasks typically involve retrieving relevant documents or knowledge elements and using them to inform downstream predictions or generations.

Key papers

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks — proposes RAG for combining retrieval with generation
• Atlas: Few-shot Learning with Retrieval Augmented Language Models — Demonstrates retrieval-augmented language models for few-shot learning on KILT benchmark
• Question Answering Systems — key application area for knowledge-intensive methods

Related topics

• Information Retrieval (technical foundation)
• Question Answering Systems (primary application)
• Fact-checking and corrections (related task)
• Retrieval-Based Methods (architecture pattern)
• Neural language models (parametric knowledge)