Lexical search retrieves content using sparse, term-based representations. In a retrieval-augmented generation (RAG) pipeline, it can complement dense embedding search as part of hybrid retrieval. The practical goal is to surface the right context for a user’s question: even an accurate document and a capable language model cannot compensate for a missed retrieval. section 3.8.12 section 4.0
Hybrid pipelines can combine dense and sparse retrieval with reranking to improve precision. The available treatment identifies that design but does not specify a lexical index, scoring method, or fusion algorithm. Chunking, retrieval quality, and continuous evaluation remain production concerns. section 4.0 section 3.8.12
Do not mistake keyword checks elsewhere in an application for a full lexical retriever. A blocked-word guardrail or keyword-based model router tests for specified strings; neither ranks documents for retrieval. Both illustrate a limitation of literal matching: synonyms, creative phrasing, or unexpectedly complex queries can defeat simple lists. section 6.4.2 section 9.2.4
