Large language models can translate without understanding meaning
Translation Indeterminacy and the Distributional Fallacy
Computation and Language
Summary
People often think language models learn meaning by looking at word patterns, but this paper argues it's the other way around: meaning in language shapes these patterns. The authors say translation between languages can still work by using patterns alone, without true understanding of meaning. They propose that real meaning comes from how people interact with the world, something current large language models don't do. So, language models can translate by mapping pattern correspondences, but they lack grounded understanding of what words actually mean.
What this means in practice
- •For machine translation engineers: Improve machine translation systems by focusing on distributional correspondences without assuming models understand meaning.
- •For human language teaching teams: Design language learning tools that highlight pattern correspondences between languages rather than relying solely on semantic understanding.
A position paper. It proposes an approach and reports no results.
Authors
Michael Carl
Abstract
Large language models (LLMs) are commonly associated with the distributional hypothesis, according to which (1) semantic meaning is grounded in distributional patterns of linguistic context, and (2) knowledge of cross-linguistic distributional correspondences allows for successful translation. This paper rejects the first claim as a causal inversion: linguistic distributions reflect patterns arising from meaning-making practices rather than constituting their source. At the same time, it accepts the second claim, arguing that translation -human or machine - can succeed without requiring access to meaning or reference. Knowledge of interlingual distributional correspondence and their inferential organization may be sufficient for translation. The paper develops an ecological-enactivist perspective, according to which reference and meaning are grounded in agent-environment interaction and stabilized through action-grounded concepts, forms of world-involving cognition that current LLMs do not possess.