Can ChatGPT Detect Intent? Evaluating Large Language Models for Spoken Language Understanding

Here we provide examples of the prompts used in the evaluation. For the cases with fewer N<30 shots, simply keep only the first N examples.

• Prompts for ChatGPT: SLURP Intent classification [0-shot], [30-shot], [20-shot, verbose], [MINDS-14 Korean 14-shot], SLURP slot filling [30-shot]
• Prompts for GPT3.5: SLURP Intent classification [30-shot], [ASR with chain-of-thought]
• Prompts for other autoregressive language models: SLURP Intent classification [30-shot]