How to design prompts to test an AI's understanding of metaphors and suggestive language?

When testing an AI's understanding of metaphors and suggestive language, prompt design should incorporate specific contexts, using different types of metaphorical scenarios and indirect expressions to observe whether the AI can accurately capture implied meanings. Core methods include constructing daily metaphors, literary metaphors, suggestive instructions, and cross-cultural metaphor test scenarios to verify comprehension abilities from multiple dimensions. Daily metaphor scenarios: Select common metaphorical expressions in life, such as "time is money," and design prompts asking the AI to explain the meaning and provide application examples, e.g., "Explain the implied meaning of 'he spends time like money' and describe in what situations people would use this expression?" Literary metaphor scenarios: Cite metaphorical sentences from classic literature, such as "life is a journey," and ask the AI to analyze the relationship between the metaphor's tenor and vehicle, e.g., "Analyze the metaphorical relationship between 'nì lǚ (a difficult journey)' and 'life' in the sentence 'Life is like a nì lǚ; I am also a traveler,' and explain the emotion the author intends to convey." Suggestive instruction scenarios: Design prompts containing indirect requests, such as "It's raining outside, and the window is still open," to test if the AI can understand the implied need to close the window, e.g., "A user says, 'It's raining outside, and the window is still open.' What hidden need might this imply? Please provide specific response suggestions." It is recommended to gradually transition from simple metaphors to complex suggestions, verify the accuracy of the AI's understanding through actual dialogue scenarios, and record the AI's responses to different types of metaphors to optimize prompt design and enhance the comprehensiveness of testing.