Chunking splits documents into segments for embedding and retrieval. Approaches include fixed character counts, sentence or paragraph boundaries, and structure-aware methods that respect document formatting such as deposition page-line format or medical record entries.
Overlap between chunks preserves context across boundaries.
Alternative Names:
Document Chunking, Text Segmentation
Why it Matters?
Chunking is the least visible and most consequential retrieval decision for legal documents. Splitting a deposition mid-answer separates the question from the response, and splitting a medical record entry from its date makes the entry unusable. Vendors applying uniform character-count chunking to litigation documents produce noticeably worse retrieval than those handling document structure, and the difference is invisible until you test on real records.
Frequently Confused with
Related terms
Frequently asked questions
Why does chunking affect accuracy?
What is structure-aware chunking?
