Difference in the Cognitive Mechanism of Predictive Processing in Computer-Mediated Communication: A Comparison Study of L2 Speakers

Published: CogSci 2025, San Francisco, CA.

This paper investigates how the cognitive machinery of next-word prediction differs between native (L1) and non-native (L2) speakers across computer-mediated communication (CMC) modalities (e.g., voice, video, text-based chat). Using LLM-approximated next-word probabilities as a proxy for human predictive processing, we model how the loss of multimodal cues (gaze, gesture, prosody) in different CMC channels disproportionately affects L2 speakers’ ability to anticipate upcoming content during turn-taking.

Findings suggest that L2 speakers rely more heavily on multimodal cues to compensate for slower lexical access, and that text-only CMC removes precisely the cues they need most—providing a cognitive-mechanistic account of why cross-linguistic conversation is harder in some CMC channels than others, and motivating channel-aware AI scaffolding.

Recommended citation: He, W. P. & Fussell, S. R. (2025). "Difference in the Cognitive Mechanism of Predictive Processing in Computer-Mediated Communication: A Comparison Study of L2 Speakers." Proceedings of the 47th Annual Meeting of the Cognitive Science Society (CogSci 2025). San Francisco, CA.
Download Paper