ガイド連動クイズ · 中 レベル

一問一答クイズ

抽出的および生成的な QA、外部証拠、検索根拠、回答境界、棄権、重複評価、言い換えテストをレビューします。

関連するガイドパス質問への回答
質問 1 の 8

抽出的質問応答システムは何をしますか?

すべての質問をプレビューする8 質問

クイズに答えると、答えと解説が表示されます。

  1. 抽出的質問応答システムは何をしますか?
    • 質問に答える文章内のテキストの正確な範囲を選択します
    • まったく新しい答えを自分の言葉で書く
    • 質問を別の言語に翻訳します
    • Web ページを人気順にランク付けします
  2. What evidence-access distinction separates closed-book QA from an open-book QA setting?
    • Open-book systems are forbidden to read documents
    • Closed-book systems must always be more accurate
    • Closed-book uses model parameters without external lookup; open-book can use supplied or retrieved documents
    • Open-book systems cannot include a language model
  3. How can retrieval-augmented generation support a QA answer?
    • By guaranteeing every retrieved passage is correct
    • By eliminating the need to interpret any passage
    • By automatically proving every output claim true
    • By providing retrieved passages as evidence for generation
  4. Which boundary scores can an extractive QA head produce when selecting an answer span?
    • Only a translation score for French output
    • Scores for possible answer-start and answer-end positions
    • Only a grammatical noun-or-verb label
    • Only the alphabetical rank of each word
  5. 文章に答えがない場合、QA システムが適切に設計されているという兆候は何ですか?
    • とにかくもっともらしい答えを発明してしまう
    • パッセージ全体を返します
    • 質問に答えられないことを認識し、そのように答えます
    • クラッシュする
  6. How does generative QA differ from requiring an extractive source span?
    • It produces answer text rather than requiring an exact span from the passage
    • It is restricted to one yes-or-no label in every task
    • It never outputs any words present in a source
    • It must output only a highlighted passage boundary
  7. In SQuAD-style QA evaluation, what can token-overlap F1 provide that exact match alone does not?
    • Proof that every claimed fact is true
    • Partial credit for answer words overlapping a reference
    • A guarantee that no answer was copied
    • The measured power usage of a server
  8. Why evaluate a QA system on paraphrased questions that preserve the intended meaning?
    • To guarantee that every new paraphrase has a different answer
    • To remove all passage evidence from every evaluation
    • To replace answer quality with filename length
    • To test sensitivity to changes in question wording