Explanatory inversion involves reformulating questions to focus on their underlying explanations or reasoning processes. By augmenting general questions with explanatory inversion, you encourage LLMs to engage in deeper reasoning and better understand cause-effect relationships. Here are several rules for creating such augmentations:

---

### **1. Convert Direct Questions into Why-Based Questions**
   - **Rule:** Reformulate the question to ask about the reasons or mechanisms behind the original query.
   - **Example:**
     - Original: "What are the symptoms of diabetes?"
     - Augmented: "Why do these symptoms occur in diabetes?"

---

### **2. Emphasize Cause-Effect Relationships**
   - **Rule:** Reframe the question to highlight causal relationships rather than surface-level descriptions.
   - **Example:**
     - Original: "What causes earthquakes?"
     - Augmented: "How do tectonic plate movements cause earthquakes?"

---

### **3. Focus on Processes or Mechanisms**
   - **Rule:** Augment questions to ask about the step-by-step processes involved in a phenomenon.
   - **Example:**
     - Original: "How does photosynthesis work?"
     - Augmented: "What are the mechanisms through which plants convert sunlight into energy during photosynthesis?"

---

### **4. Include Counterfactual Scenarios**
   - **Rule:** Reformulate the question to explore what would happen if a certain condition were different.
   - **Example:**
     - Original: "What happens if a vaccine is not administered?"
     - Augmented: "How would the spread of disease change if vaccines were unavailable?"

---

### **5. Highlight Comparisons and Contrasts**
   - **Rule:** Augment the question to compare different cases or conditions.
   - **Example:**
     - Original: "What are the effects of climate change?"
     - Augmented: "How do the effects of climate change differ between polar regions and equatorial regions?"

---

### **6. Encourage Hypothesis Exploration**
   - **Rule:** Reformulate the question to ask about possible explanations or hypotheses.
   - **Example:**
     - Original: "What is dark matter?"
     - Augmented: "What hypotheses explain the existence of dark matter?"

---

### **7. Incorporate Real-World Scenarios**
   - **Rule:** Tie the question to practical or observable scenarios to encourage applied reasoning.
   - **Example:**
     - Original: "What is the role of enzymes in digestion?"
     - Augmented: "How would the human body be affected if enzymes were absent in the digestive process?"

---

### **8. Use Chain-of-Reasoning Prompts**
   - **Rule:** Reformulate the question to require multi-step reasoning to reach the answer.
   - **Example:**
     - Original: "What are the main causes of deforestation?"
     - Augmented: "What are the social and economic factors that lead to deforestation, and how do they interact?"

---

### **9. Introduce Temporal Dynamics**
   - **Rule:** Reformulate questions to explore how phenomena change over time.
   - **Example:**
     - Original: "What happens during a solar eclipse?"
     - Augmented: "How does the alignment of celestial bodies change over time during a solar eclipse?"

---

### **10. Integrate Explanatory Questions**
   - **Rule:** Directly ask for an explanation of the reasoning behind an answer.
   - **Example:**
     - Original: "What is the function of the heart?"
     - Augmented: "How does the heart's structure enable its function of pumping blood?"

---

### Implementation in LLM Training
By applying these rules to augment questions, you create a dataset that promotes deeper reasoning and explanatory learning for LLMs. These augmented questions encourage the model to provide richer, more detailed, and logically consistent answers.