Question Generation is a pre-training or auxiliary task where the model is trained to generate a valid specific question given a passage and an answer — turning the standard QA task around (Answer → Question) to improve the model's understanding of the relationship between information and inquiries.

Structure

• Input: "Context: Paris is the capital of France. Answer: France."
• Output: "What country is Paris the capital of?"
• Usage: Used to synthesize data for QA training or as a pre-training objective (e.g., in T5).
• Consistency: Can act as a consistency check — does the generated question lead back to the answer?

Why It Matters

• Data Augmentation: Can generate infinite QA pairs from raw text to train QA models.
• Dual Learning: Training on both Q→A and A→Q improves performance on both.
• Reading Comprehension: forces the model to understand what simple facts can answer.

Question Generation is playing Jeopardy — giving the answer and asking the model to come up with the correct question.