Cross-lingual LLM-mediated affective communication is bounded in fidelity and shaped by prosocial emotion sensitivity

Scalable affective assessment requires generative stimuli that can be tailored to diverse populations, but two prerequisites for individualised use remain untested: whether LLM-mediated affective communication guarantees fidelity of transfer across linguistic context, and whether recipient psychopathology modulates the response to LLM-generated affective stimuli. We introduce a pipeline quantifying the fidelity of LLM-mediated affective communication and testing whether psychopathology modulates responses to LLM-generated stimuli. An English-speaking source cohort (n = 547) provided core affect ratings (valence, arousal, focus, dominance) for 18 emotion constructs.

Conditioned on these ratings, GPT-4o generated 180 text and 18 image stimuli, rated by a Korean-speaking evaluation cohort (n = 91) along the same dimensions. Cross-lingual alignment was quantified as cosine similarity and benchmarked against chance, a within-cohort semantic-distinctness threshold (ε), and another same-language cohort (n = 76); associations with six psychopathology measures were tested using linear mixed-effects models. Similarities exceeded chance-level across both formats but did not reach ε, and same-language alignment exceeded cross-lingual alignment.

After multivariate adjustment, sensitivity to prosocial emotions (PFQ-2) was the sole predictor of lower-magnitude and more variable responses. LLM-mediated affective communication is bounded and selectively shaped by reactivity to prosocial emotions, providing a measurable channel for individualised affective assessment. This research was supported by the Si-Hyung Lee Social Psychiatry Research Grant from the Korean Neuropsychiatric Research Foundation in 2025 and by the Medical Scientist Training Program and the Basic Research Program (RS-2025-00562004) funded by the Ministry of Science & ICT of Korea.

The funders had no role in the study design, data collection, analysis, interpretation or writing of the manuscript. The Korean Neuropsychiatric Research Foundation,the Ministry of Science & ICT of Korea (RS-2025-00562004). Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea Severance Hospital, Yonsei University, Seoul, Republic of Korea College of Medicine, Chungbuk National University, Cheongju-Si, Republic of Korea Graduate School of Medical Science and Engineering, Korea Advanced Institute of Science and Technology, R102, 193 Munji-Ro, Yuseong-Gu, 34051, Daejeon, Republic of Korea JK and BJ have a patent pending, assigned to Korea Advanced Institute of Science and Technology.

The remaining authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.

You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.