question answering
/ KWES-chun AN-ser-ing /
Question answering is exactly what it sounds like: you ask a question in plain language and the system gives you a direct answer, not a list of links to wade through. "How tall is Mount Everest?" should come back with "about 8,849 meters," not ten web pages. It is the difference between a search engine that hands you a haystack and an assistant that finds the needle.
Two broad styles exist. Extractive QA is given a passage and must find the span of text inside it that answers the question — like a reading-comprehension quiz where every answer is somewhere in the paragraph. Open-domain QA is harder: the system must first retrieve relevant documents from a huge collection and then extract or compose an answer, the recipe behind "retrieval-augmented" assistants. Generative QA, used by chatbots, writes the answer in its own words, drawing on what it learned during training.
QA is the backbone of voice assistants, customer-support bots, and search boxes that now reply in sentences. But the honest limits are sharp. A system can answer confidently and be completely wrong, especially when it generates from memory rather than looking something up — it has no built-in sense of "I don't actually know this." It can be fooled by how a question is phrased, struggle with multi-step reasoning, and parrot outdated facts. Grounding answers in a trusted source and showing where they came from is the current best defense.
Passage: "The Eiffel Tower, completed in 1889, stands 330 meters tall." Question: "When was the Eiffel Tower completed?" Extractive QA answers by pointing to the exact span "1889" — the answer is found, not invented.
Extractive QA locates the answer inside a given passage rather than generating it.
A QA system has no built-in sense of "I don't know" — it can answer wrong with total confidence, especially when generating from memory. Grounding answers in a cited source is the main safeguard.