Large Language Models process instructions as social acts rather than technical specifications, making 'imperative mood' prompts behave inconsistently across different languages.
March 27, 2026
Original Paper
Imperative Interference: Social Register Shapes Instruction Topology in Large Language Models
arXiv · 2603.25015
The Takeaway
The paper shows that English and Spanish handle authority differently; rewriting imperative commands ('NEVER do X') into declarative facts ('X: disabled') reduces cross-linguistic variance by 81%. This has major implications for global alignment and consistent system prompt design.
From the abstract
System prompt instructions that cooperate in English compete in Spanish, with the same semantic content, but opposite interaction topology. We present instruction-level ablation experiments across four languages and four models showing that this topology inversion is mediated by social register: the imperative mood carries different obligatory force across speech communities, and models trained on multilingual data have learned these conventions. Declarative rewriting of a single instruction blo