Anthropomorphic perception and meaning-making in human–AI interaction: a comparison of mixed reality and ChatGPT using multimodal social semiotics and LIWC

Abstract This study investigates how students’ interactions with embodied and disembodied generative conversational AI in an English for Specific Purposes classroom indicate anthropomorphism and meaning making during interaction. Fifty ESL students interacted with two different modalities of generative AI for six weeks: an embodied conversational agent (ECA) in mixed reality (MR) and ChatGPT voice applications. The study employed multimodal social semiotics (MSS) and computational linguistic analysis (LIWC) to examine both observable multimodal conduct (gestures, gaze, and spatial orientation) and linguistic markers of social orientation. MSS findings revealed higher modal density in the MR condition, with learners demonstrating sustained gaze, forward lean, and spontaneous co-speech gestures, positioning the agent as socially accountable. LIWC analysis showed that ChatGPT interactions contained significantly higher social processes, affiliation language, and self-referential pronoun use, whereas MR interactions demonstrated higher emotional tone. The results indicate that anthropomorphism operates as both an embodied semiotic practice and a linguistically enacted social stance, shaped by interactional affordances and technological modality.