France or Spain or Germany or France: A Neural Account of Non-Redundant Redundant Disjunctions

TL;DR

This paper investigates how neural mechanisms in language models explain the acceptance of seemingly redundant disjunctions in context, combining behavioral evidence with neural modeling.

Contribution

It introduces a neural account for non-redundant disjunctions, highlighting mechanisms like binding and attention in Transformer models, complementing symbolic approaches.

Findings

Humans and language models accept non-redundant disjunctions across contexts.

Redundancy avoidance in models arises from binding and attention mechanisms.

Neural explanations complement symbolic analyses of semantic interpretation.

Abstract

Sentences like "She will go to France or Spain, or perhaps to Germany or France." appear formally redundant, yet become acceptable in contexts such as "Mary will go to a philosophy program in France or Spain, or a mathematics program in Germany or France." While this phenomenon has typically been analyzed using symbolic formal representations, we aim to provide an account grounded in artificial neural mechanisms. We first present new behavioral evidence from humans and large language models demonstrating the robustness of this apparent non-redundancy across contexts. We then show that, in language models, redundancy avoidance arises from two interacting mechanisms: models learn to bind contextually relevant information to repeated lexical items, and Transformer induction heads selectively attend to these context-licensed representations. We argue that this neural explanation sheds light on the mechanisms underlying context-sensitive semantic interpretation, and that it complements existing symbolic analyses.