Question Answering
Getting a machine to answer a question in natural language — the task that quietly turned from "find the passage" into "generate the answer," and defines how we use AI today.
Reviewed July 15, 2026Stable
Reach for something else instead
- Extractive QA when you need the answer grounded in a specific passage.
- Semantic search when the user wants relevant documents, not a synthesised answer.
- RAG as the standard bridge — retrieve, then generate from what was retrieved.
Further reading
- Rajpurkar et al. (2016), SQuAD: 100,000+ Questions for Machine Comprehension — the benchmark that defined extractive QA.
- Chen et al. (2017), Reading Wikipedia to Answer Open-Domain Questions — the retriever-reader architecture behind RAG.
- Lewis et al. (2020), Retrieval-Augmented Generation — grounding generative QA in retrieved documents.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Using generative QA where answers must be verifiable, and getting confident hallucinations.
- Blaming the model for wrong answers when the retrieval step failed to surface the right passage.
- Evaluating open-ended answers by exact string match, which misses correct paraphrases.
At a glance
FieldLanguage & LLMs
Two modesextractive (find the span) vs. generative (compose the answer)
Grounding fixRAG
DifficultyIntermediate
Where this sits
A destination. 1 concept lead here, and nothing in the corpus depends on it.
1Levelsteps in
1Needs firstconcepts
0Opens upnothing further
1Areastays here
Learn these firstNatural Language Processing
Computed from the prerequisite graph, not assigned. How this works