Meaning without reference in large language models
Steven T. Piantadosi Department of Psychology Helen Wills Neuroscience Institute University of California, Berkeley \AndFelix Hill DeepMind
Abstract
The widespread success of large language models (LLMs) has been met with skepticism that they possess anything like human concepts or meanings. Contrary to claims that LLMs possess no meaning whatsoever, we argue that they likely capture important aspects of meaning, and moreover work in a way that approximates a compelling account of human cognition in which meaning arises from conceptual role. Because conceptual role is defined by the relationships between internal representational states, meaning cannot be determined from a model’s architecture, training data, or objective function, but only by examination of how its internal states relate to each other. This approach may clarify why and how LLMs are so successful and suggest how they can be made more human-like.
中文速览
大型语言模型（LLM）到底有没有"意义"或"理解"，一直是学界争议的焦点，批评者认为仅凭文本预测训练、缺乏与真实世界对象的指称（reference）连接，模型不可能获得真正的意义。这篇文章反驳了这一论断，指出哲学和认知科学早已发现"意义由指称决定"的观点存在根本缺陷——无论是"正义""虚构概念"还是"邮票"这类具体词汇，其意义的核心都不在于它指向哪个物体，而在于它与其他概念之间的关系网络，这正是哲学中"概念角色理论"（conceptual role theory）的核心主张。作者据此论证，LLM 在训练过程中所形成的内部表征之间的关系结构，已经在很大程度上近似于这种人类意义的工作方式——神经影像研究甚至表明，训练数据越多的 LLM，其表征几何结构与人类大脑语义加工的匹配度越高。这一视角不仅有助于解释为何 LLM 能在推理、叙事、问答等复杂任务上取得成功，也为未来改进方向提供了指引：与其纠结于模型是否"有指称"，不如持续丰富其概念关系网络，正如人类对"水"的理解在发现H₂O之后得到深化，但从未经历从"无意义"到"有意义"的骤变。
原文 arXiv:2208.02957;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2208.02957v2