What are the criteria for question extraction?

While there isn't a single, universally agreed-upon set of criteria, here are some key considerations for effective question extraction:

I. Purpose and Context:

• Goal-Oriented: Questions should be relevant to a specific goal or task. Are you trying to assess understanding, trigger a process, provide information, or generate discussion?
• Domain-Specific: The subject matter (medical, legal, technical, etc.) will greatly influence the types of questions that are appropriate.
• Audience: Who are the questions intended for? (Experts, novices, general public)

II. Linguistic Features:

• Interrogative Words: The presence of words like "who," "what," "where," "when," "why," "how," "is," "are," "do," "does," "can," "could," "should," "would," "will," etc., often signals a question.
• Question Marks: A question mark (?) at the end of a sentence is a strong indicator. But be careful: A question mark doesn't always mean it's a question you want to extract (e.g., rhetorical questions).
• Sentence Structure: Inverted subject-verb order (e.g., "Is he going?") is common in questions. However, many questions have standard subject-verb order.
• Modal Verbs: Modal verbs often signal questions ("Could you help me?", "Should I do this?").

III. Semantic Considerations:

• Information Seeking: Questions should express a genuine desire to know something.
• Clarity and Completeness: Extracted questions should be understandable without significant context from the surrounding text. They should make sense on their own.
• Focus: Questions should be focused on a single, identifiable topic. Avoid questions that are overly broad or vague.
• Relevance: The question should be relevant to the core theme or themes of the document.

IV. Avoiding Common Pitfalls:

• Rhetorical Questions: These are statements disguised as questions and don't require an answer. Exclude them if you're looking for information-seeking questions. (Example: "Who cares?")
• Incomplete Questions: If the question is fragmented and lacks crucial information, it might not be useful on its own.
• Embedded Questions: Sentences that contain questions but are not themselves interrogative. (Example: "I wonder what time it is.") Decide if the embedded question is the primary focus.
• Indirect Questions: Questions phrased as requests or suggestions. (Example: "Could you tell me the time?") Consider extracting the core question ("What time is it?")
• Titles and Headings: Titles formatted as questions might be misleading. Assess whether they represent a real question or just a topic.
• Questions Used as Examples: Questions that are used within a text to illustrate a particular concept, rather than to seek information. (Example: "Consider the question, 'What is the meaning of life?'")

In summary, good question extraction aims to identify sentences that are explicitly interrogative, semantically clear, relevant to the overall context, and express a genuine need for information or action. The specific criteria will depend on the purpose of the question extraction process.