# Chunking

> Chunking is splitting documents into smaller passages before indexing them for retrieval, so that each retrieved unit is small enough to fit the context and focused enough to match a query.

- Identifier: PTL-0125
- Category: Retrieval & Tool Use
- Canonical URL: https://protologue.com/t/chunking/
- Also known as: document chunking, text splitting, passage segmentation, late chunking

## Description

Chunk size and overlap are the main levers. Chunks that are too large dilute the embedding and waste context, while chunks that are too small lose the surrounding meaning. Common strategies split by fixed token counts, by document structure such as headings and paragraphs, or by meaning. Research has found that retrieving finer units, such as self-contained propositions, can improve retrieval, and approaches such as contextual retrieval and late chunking restore context that splitting removes.

## Related terms

- [Embedding](https://protologue.com/t/embedding/)
- [Contextual Retrieval](https://protologue.com/t/contextual-retrieval/)
- [Retrieval-Augmented Generation](https://protologue.com/t/retrieval-augmented-generation/)
- [Dense Retrieval](https://protologue.com/t/dense-retrieval/)

## Sources

- Chen et al. (2023). Dense X Retrieval: What Retrieval Granularity Should We Use?. https://arxiv.org/abs/2312.06648
- Günther et al. (2024). Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding Models. https://arxiv.org/abs/2409.04701

## Cite this entry

Protologue. (2026). Chunking. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0125). https://protologue.com/t/chunking/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)