Source-linked AI summary
Meaning without reference in large language models
Steven T. Piantadosi, Felix Hill
TL;DR
Against the claim that text-trained LLMs cannot have meaning without reference, the paper argues that meaning can arise from conceptual roles—the relations among internal states. It concludes that LLMs likely capture important aspects of meaning, though multimodal models still lack parts of a complete semantics.
Problem
Critics argue that text-trained LLMs lack meaning because they lack reference, but the paper challenges whether reference determines meaning.
Source: p. 1Method
The paper applies conceptual role theory, locating meaning in relationships among internal representational states rather than architecture, training data, or objective function.
Source: pp. 3–4Results
LLMs likely capture important aspects of meaning: their relational geometry increasingly reflects human data, and their competencies require sensitivity to conceptual roles.
Source: pp. 3, 5Takeaways & Limitations
Reference is one aspect of a word’s full conceptual role, relevant for some concepts but not others.
Source: p. 4Even multimodal LLMs lack aspects of a complete theory of semantics, including simulating how physical or linguistic behavior affects their environment and representing users’ goals and desires.
Source: pp. 3–4
Abstract
from arXiv · showhide
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.