Large language models use culture and practice to do language
Communication, Vocalisation, and Language
Computation and Language
Summary
The paper argues that large language models (LLMs) don't truly understand language like humans do. Instead of forming mental images or ideas of words, these models learn language by capturing shared cultural behaviors and social practices. This way, LLMs perform language tasks by imitating the way people use symbols within their communities. The authors suggest this is how these models 'do' language without deep comprehension.
What this means in practice
- •For natural language engineers: Design language systems that improve performance by focusing on cultural context and social practices instead of explicit symbol representations.
- •For sociolinguistics software developers: Develop tools that model human communication by integrating cultural practices, inspired by LLM behavior described in this paper.
A position paper. It proposes an approach and reports no results.
Authors
Peter Wallis
Abstract
The contention is that the current batch of Large Language Models don't *understand* language, but they *do* it. Rather than LLMs forming some kind of ``distributed representation'' of symbols, the claim is that LLMs are capturing ``Cultural Intelligence'' and grounding symbols in what the sociologists call practices.