Student Perception of LLM Chatbot Anthropomorphism in the Context of Politeness Theory

Abstract

Introduction . The phenomenon of anthropomorphism acquires particular pragmatic significance in the modern context, especially in connection with the development and active implementation of Large Language Models (LLMs) in social practices. Today education is one of the areas where LMMs are being implemented most actively. This study aims to examine the phenomenon of anthropomorphism in students’ perception of chatbots in the context of politeness theory. Materials and Methods . An adapted version of the Godspeed Questionnaire Series was used to measure anthropomorphism, animism, and perceived intelligence. The communication strategies of chatbots were designed based on P. Brown and S. Levinson’s Politeness theory. The experiment involved 543 students (aged 17–20) from three Russian universities. Three chatbots were developed specifically for the experiment: Nomi (face-saving), Vector (face-threatening), and Sero (neutral). The tests were conducted in three communication frames: explanation, feedback, and recommendation; each interaction between a student and a chatbot lasted no more than 10 minutes. Results . Vector received the highest average anthropomorphism score (M = 3.123) compared to Sero (M = 2.598) and Nomi (M = 2.924). It was found that the perception of chatbots is stratified: Vector has a high sense of human-likeness combined with lower scores for animacy and intelligence. It is noteworthy that the feedback frame enhances the anthropomorphization of chatbots, while recommendations do not. No significant correlations were found with AI usage experience. Discussion and Conclusion . Anthropomorphism is a multidimensional and context-dependent phenomenon. The active use of AI technologies in everyday life leads to a partial dissociation of previously closely related concepts: human likeness, animation, and perceived rationality, as well as the transformation of some social interaction frames where digital partners actively integrate.

Ähnliche Arbeiten