Cell-Based Representation of Relational Binding in Language Models

TL;DR

This paper reveals that Large Language Models encode relational binding through a low-dimensional Cell-based Binding Representation (CBR), which can be decoded, manipulated, and causally linked to model performance.

Contribution

The study introduces the CBR as a linear subspace encoding entity-relation bindings in LLMs, with methods for decoding, transferring, and manipulating these representations.

Findings

CBR forms a grid-like geometry in activation space.

Decoding CBR indices is linearly feasible across domains.

Manipulating CBR alters relational predictions causally.

Abstract

Understanding a discourse requires tracking entities and the relations that hold between them. While Large Language Models (LLMs) perform well on relational reasoning, the mechanism by which they bind entities, relations, and attributes remains unclear. We study discourse-level relational binding and show that LLMs encode it via a Cell-based Binding Representation (CBR): a low-dimensional linear subspace in which each ``cell'' corresponds to an entity--relation index pair, and bound attributes are retrieved from the corresponding cell during inference. Using controlled multi-sentence data annotated with entity and relation indices, we identify the CBR subspace by decoding these indices from attribute-token activations with Partial Least Squares regression. Across domains and two model families, the indices are linearly decodable and form a grid-like geometry in the projected space. We further find that context-specific CBR representations are related by translation vectors in activation space, enabling cross-context transfer. Finally, activation patching shows that manipulating this subspace systematically changes relational predictions and that perturbing it disrupts performance, providing causal evidence that LLMs rely on CBR for relational binding.