The Semantic Hub Hypothesis: Language Models Share Semantic Representations Across Languages and Modalities

TL;DR

This paper proposes the semantic hub hypothesis, suggesting that language models develop a shared, integrated semantic representation space across languages and modalities, which is actively used during processing.

Contribution

It introduces the semantic hub hypothesis and provides evidence that models learn a shared semantic space across languages and modalities, extending the hub-and-spoke model from neuroscience.

Findings

Model representations for equivalent inputs in different languages are similar.

Shared semantic space extends to arithmetic, code, and visual/audio data.

Interventions in the shared space affect outputs across data types.

Abstract

Modern language models can process inputs across diverse languages and modalities. We hypothesize that models acquire this capability through learning a shared representation space across heterogeneous data types (e.g., different languages and modalities), which places semantically similar inputs near one another, even if they are from different modalities/languages. We term this the semantic hub hypothesis, following the hub-and-spoke model from neuroscience (Patterson et al., 2007) which posits that semantic knowledge in the human brain is organized through a transmodal semantic "hub" which integrates information from various modality-specific "spokes" regions. We first show that model representations for semantically equivalent inputs in different languages are similar in the intermediate layers, and that this space can be interpreted using the model's dominant pretraining language via the logit lens. This tendency extends to other data types, including arithmetic expressions, code, and visual/audio inputs. Interventions in the shared representation space in one data type also predictably affect model outputs in other data types, suggesting that this shared representations space is not simply a vestigial byproduct of large-scale training on broad data, but something that is actively utilized by the model during input processing.