Using Prompt-based AI to Identify Miratives in Constructions

Despite the availability of many sophisticated machine learning models for sentiment analysis and emotion recognition, few are specifically designed to detect verbal expressions of surprise or unexpectedness. The use of prompting, using natural language to direct Large Language Model (LLM) outputs, allows non-AI experts, including linguists, to overcome the time- and effort-consuming process of programming and training models. This research investigates the potential for identifying miratives (expressions of surprise or unexpectedness) in Chinese and English constructions through prompt-based AI with the pretrained LLM GPT-3. We apply the proposed method to a corpus recognition experiment with a dataset comprising 100 sentences in both Chinese and English. Miratives in the dataset are identified using zero-shot and few-shot prompts, with zero-shot prompting serving as the baseline model. A fine-tuned LLM model is also performed on the dataset. Additionally, a comparative analysis of the performance of the prompt-based AI models and the fine-tuned LLM model on the recognition task is performed. The results show which model achieves the most satisfactory performance on the dataset. The obtained results are then utilized to assess the effectiveness of the prospective method in identifying miratives in linguistic constructions. Our tentative conclusion is that if prompts are adequately designed, the recognition efficiency of prompt-based AI models could be somewhat different from fine-tuned LLM, but the former would be more resource- and time-saving and would attract more linguists to use prompt-based AI.