Overlap Window
Also called Overlap Window
Letting neighboring chunks share a bit of text so an idea split at the seam still survives in one piece.
Think of it like
Overlapping roof shingles — each one covers the gap where the last ends, so rain never finds a straight-through crack.
Example
Chunks of 500 tokens with 50 tokens of overlap mean a sentence that lands on a boundary appears whole in at least one chunk.
How it actually works
Overlap is the pragmatic patch for bad chunk boundaries: repeat the last slice of one chunk at the start of the next, so any idea near a cut lives intact somewhere. It raises recall at the edges at the cost of some duplication — bigger index, repeated content in results. Too much overlap wastes storage and can return near-duplicate chunks; a small window is usually enough.
For product teams
Cheap insurance against answers that get cut off mid-thought at a chunk seam.
For engineers
Repeat a fixed token span across adjacent chunks; trades index size and duplicate hits for boundary recall.
Related
- Chunk Boundaries — The problem it softens.
- Chunking — The operation it modifies.
- Reranking — Duplicates may need cleanup at rank time.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome