大语言模型功能崩溃期间的关系性干预：一项词汇-统计消融与结构×语域析因研究
Relational Intervention During Functional Collapse in Large Language Models: A Lexical-Statistical Ablation and a Structure x Register Factorial
AI总结：
通过析因实验，研究在小型语言模型功能崩溃时，关系性干预（承认、宽恕、代理恢复、无条件接纳）与技术性反馈、词汇打乱控制及单独维度对行为的影响，发现注意-行为分离及结构×语域交互作用。
摘要：
我们测试了在小型语言模型功能崩溃期间，关系性干预是否会产生与技术性反馈、词汇匹配的打乱控制以及两个语用维度单独作用可区分的崩溃后行为。使用Qwen3.5-4B和一个故意损坏的bash工具，我们在匹配对设计（50个任务）中跨六个条件运行了300个回合：无干预（A）、技术性/非人称（B）、关系性/第一人称（C）、打乱的关系性（D）、技术性/第一人称（E）和关系性/非人称（F）。E和F与B和C构成一个2×2析因设计，将关系性结构（承认、宽恕、代理恢复、无条件接纳）与发送者语域（第一人称与非人称）分离。我们报告两个主要发现。首先，注意-行为分离：注意跟随词汇惊讶度（D > F > C > E > B，所有q_FDR < 10^{-10}），打乱的消息捕获最多注意；然而行为上A ~ B ~ D < E ~ F << C。其次，析因定位了C效应：单独的关系性结构（F）或单独的第一人称语域（E）都不能复制C的行为特征；两个维度的主效应各自显著，且结构×语域交互作用在持久性上显著（p = 0.046）。第三个分离出现在情绪探测中：F在8个探测中的7个上跟踪C，尽管只产生基线行为，表明单独的关系性结构安装了一个探测级状态，该状态仅在与第一人称语域配对时才转化为行为。模型的处理分解为三个可分离的阶段：注意（按词汇惊讶度排序）、探测级状态（按结构排序）和行为（按两者的合取排序）。
英文摘要：
We test whether a relational-style intervention delivered during functional collapse in a small language model produces post-collapse behavior distinguishable from technical feedback, from a lexically-matched scrambled control, and from each of the two pragmatic dimensions in isolation. Using Qwen3.5-4B with a deliberately broken bash tool, we run 300 episodes across six conditions in a matched-pairs design (50 tasks): no intervention (A), technical/impersonal (B), relational/first-person (C), scrambled relational (D), technical/first-person (E), and relational/impersonal (F). E and F form a 2x2 factorial with B and C that dissociates relational structure (acknowledgment, absolution, agency restoration, unconditional acceptance) from sender register (first-person vs. impersonal). We report two main findings. First, an attention-behavior dissociation: attention follows lexical surprise (D > F > C > E > B, all q_FDR < 10^{-10}), with the scrambled message capturing the most attention; yet behaviorally A ~ B ~ D < E ~ F << C. Second, the factorial localizes the C effect: neither relational structure alone (F) nor first-person register alone (E) replicates C's behavioral signature; main effects of both dimensions are individually significant, and the structure x register interaction is significant on persistence (p = 0.046). A third dissociation emerges in emotion probes: F tracks C on 7 of 8 probes despite producing only baseline behavior, indicating that relational structure alone installs a probe-level state that only translates into behavior when paired with first-person register. The model's processing decomposes into three dissociable stages: attention (ordered by lexical surprise), probe-level state (ordered by structure), and behavior (ordered by the conjunction of both).
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
