顺应与建构：ChatGPT的话语调适与人机关系的协同演化

Between Adaptation and Co-Construction: ChatGPT’s Discursive Alignment and the Evolving Human-Machine Relationship

摘要: 大型语言模型ChatGPT在实际交互中表现出显著的输出调适现象。本研究深入考察了ChatGPT的话语调适策略及用户在此情境下的技术实践。研究发现，ChatGPT主要采用基于事实的调适与基于情感的调适两种策略来响应用户。用户往往出于情感需求等原因主动接纳并利用ChatGPT的调适特性，通过提示词对其进行引导，并在交往过程中产生反向学习效应，由此形成一种相互建构、互为参照的人机关系。话语调适机制削弱了人机交流中异质他者的在场感，使交互更多呈现为自我投射的回响而非真正的共鸣，交往的服务化取向推动人机关系走向可计算、可预测的标准化模式。在此背景下，人保持主体性需要在技术使用中保持反思性距离，并回归真实的人际交往与生活经验以寻求实质性共鸣。

Abstract: Large language model ChatGPT exhibits significant output adaptation phenomena in actual interactions. This study conducts an in-depth investigation into ChatGPT’s discourse adaptation strategies and users’ technological practices in this context. The findings reveal that ChatGPT primarily employs two adaptation strategies—fact-based adaptation and emotion-based adaptation—to respond to users. Users, often driven by emotional needs, actively embrace and leverage ChatGPT’s adaptive features, guiding it through prompts and generating a reverse learning effect in the process of interaction. This gives rise to a mutually constructive, mutually referential human-machine relationship. The discourse adaptation mechanism weakens the presence of the heterogeneous other in human-machine communication, rendering the interaction more as an echo of self-projection than genuine resonance. The service-oriented orientation of interaction propels human-machine relations toward a calculable, predictable, and standardized mode. In this context, maintaining human agency requires sustaining a reflexive distance in technology use and returning to authentic interpersonal communication and lived experience to seek substantive resonance.