A Pragmatic Study of Indirectness in Artificial Intelligence’s Generated Language

Abstract

Artificial intelligence tries mainly to create human-like language as means of communication. Since the creation of AI powered systems, such as chatbots and virtual assistant, their language become more advanced. However, the extent to which degree AI’s language can imitate human’s language pragmatically is still a question needs to be answered. Thus, this study detects the ability of different AI assistants to produce indirect language. 50 questions of various types were asked to different AI assistants like DeepSeek, Gemini and Chat GPT. The results show that all the three programs use indirect language strategies and devices, yet, they vary in the usage. ChatGPT’s language is more indirect than Gemini and Deepseek. It uses assertives qualified with epistemic hedges and exhibits negative politeness as well as different linguistic markers. In contrast, DeepSeek’s language is direct and authorative as it uses clear assertive language and limited use of hedges. Gemini occupied a middle position for its indirectness remained less sophisticated and less frequent than that of ChatGPT.

Ähnliche Arbeiten