Natural-language querying retrieves or ranks information from ordinary human-language requests rather than relying only on fixed keywords. Semantic search and question-answering implement it with dense vectors. section 15.3.1
A request such as “shirt without stripes” expresses a constraint that a basic inverted index may miss. Dense-vector retrieval improves query and content understanding, while large language models can interpret instructions and synthesize information. section 15.3.1
Practical search systems can combine natural-language querying with:
- Traditional inverted indexes for established keyword retrieval. section 15.3.1
- Knowledge graphs and semantic functions for manually supported query interpretation. section 15.3.1
- Signals boosting, learning to rank, and click models for relevance optimization. section 15.3.1
Dense vectors are unlikely to replace every established search technique. Hybrid systems instead combine vector retrieval, indexes, graphs, and ranking signals to improve content and query understanding. section 15.3.1
