question generation

**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.