Skip to content
BENIBIX Learn

Question Answering

AI🔤 NLP & LanguageLanguage Tasks

Also known as Question Answering System, Extractive Question Answering, Open-Domain Question Answering

Giving a direct answer to a question in plain words. Not a list of links.

Think of it like

It is like asking a smart friend a question. You get the answer, not a pile of books.

Example

Jack asks when the store closes on Sunday. The model reads the store's page and says 5 pm. It points to the line that says so.

Why it matters

People want an answer, not a stack of pages. This task turns a pile of text into one reply.

Definition

Question answering is the task of giving a direct answer to a question asked in plain words. In extractive QA, the model is given a passage and points to the words in it that answer the question. In open QA, it must find the answer itself, from what it learned or from a search. Many assistants and search pages now give answers like this at the top. A good system also says when the answer is not there.
Right: question answering finds the one line that answers, and gives just that.

Common mistake

Expecting an answer when the text has none. A good system says it was not found instead of guessing.

Key takeaways

  1. 1
    An answer, not links. One direct reply.
  2. 2
    Point or find. From a passage, or from a search.
  3. 3
    Know when to stop. Say when it is not there.