Late interaction – Definition & Meaning

Definition of late interaction

: a neural retrieval architecture that encodes queries and documents separately but delays their detailed comparison until search time, when token-level representations are matched to produce a relevance score
- The search engine used late interaction so that document vectors could be prepared in advance while query tokens still received fine-grained matches.
- Late-interaction retrieval found a passage whose overall embedding looked ordinary but whose individual terms closely matched the question.

Origin & history

first known use2020

The label was established by the 2020 ColBERT paper, which introduced contextualized late interaction over BERT representations. ColBERT stores a set of token-level vectors for each document and uses a MaxSim operation to compare them with query-token vectors.

Test yourself

Which of these is the meaning of late interaction?

Related terms

embedding a list of numbers (a vector) that represents the meaning of a piece of text, an image, or… semantic search search that matches the meaning of a query rather than its exact words, usually by comparing… reranking a second step in search where a more careful model re-scores the top results so the most… vector database a database built to store embeddings and quickly find the ones most similar to a query, used… retrieval-augmented generation a technique in which a system first searches a collection of documents for relevant passages…

Cite this entry

"Late interaction." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/late-interaction/. Accessed 8 Oct. 2026.