拉里让车停下，但模型没有注意到：Transformer对M-启发式的盲区
Larry Caused the Car to Stop, But the Model Didn't Notice: Transformer Blindness to the M-Heuristic
查看机构详情
- University of Bucharest（布加勒斯特大学）
机构由 AI 辅助整理，请以论文原文为准。
浏览论文内容
中文总结 AI 辅助
本文研究Transformer模型是否运用M-启发式进行语用推理，通过对比词汇与迂说使役结构，发现DeBERTa等模型未捕捉语用差异，但指令可激活该原则。
中文摘要 AI 辅助
现代Transformer模型通过句子嵌入在捕捉语义关系方面表现出色，但其进行语用推理的能力仍未得到充分研究。本文研究基于编码器的Transformer（如DeBERTa）是否采用M-启发式（新格赖斯原则，即标记性语言形式蕴含标记性意义）。我们通过对比词汇使役结构（如“拉里停下了车”）与迂说使役结构（如“拉里使车停下”），并使用自然语言推理框架来检验这一假设。我们在涉及15个双及物动词的188种条件下进行的实验表明，DeBERTa、RoBERTa和BART均未表现出捕捉这些形式之间语用差异的证据，其中DeBERTa对100%的情况预测为“中性”。探针分析最初表明存在表征-使用分离，但对照实验揭示探针追踪的是句法复杂性，而非使役语用。对30个三元组的语义相似性分析显示，在29/30的情况下，迂说使役结构与无中介的方式描述更为接近，这与嵌入空间中M-启发式的预测相反。在明确的元语言框架下，Gemini Flash-Lite在逐项痕迹上达到100%的准确率，因此该原则在指令下可用，但在默认的NLI中未被使用。
英文摘要
Modern transformer models excel at capturing semantic relationships through sentence embeddings, yet their ability to perform pragmatic reasoning remains understudied. This paper investigates whether encoder-based transformers such as DeBERTa employ the M-Heuristic (the neo-Gricean principle that marked linguistic forms implicate marked meanings). We test this hypothesis by contrasting lexical causatives (e.g., ``Larry stopped the car'') with periphrastic causatives (e.g., ``Larry caused the car to stop'') using a Natural Language Inference framework. Our experiments across 188 conditions with 15 ambitransitive verbs reveal that DeBERTa, RoBERTa, and BART show no evidence of capturing the pragmatic distinction between these forms, with DeBERTa predicting ``Neutral'' for 100% of cases. Probing analysis initially suggested a representation-use dissociation, but control experiments reveal the probe was tracking syntactic complexity, not causative pragmatics. Semantic similarity over 30 triplets places periphrastic causatives closer to unmediated manner descriptions in 29/30 cases, opposite to M-Heuristic predictions in the embedding space. Under explicit metalinguistic framing, Gemini Flash-Lite reaches 100% with item-specific traces, so the principle is available under instruction yet unused in default NLI.
