cs.CLNov 17, 2026

Polarity Composition Hypothesis: Causal Evidence from Negation in NLI

Authors: Nhi Hoai Doan, Tatsuya Hiraoka, Munachiso Samuel Nwadike, Kentaro Inui

Organizations: Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) · RIKEN

Abstract

We investigate how language models process negation in natural language inference (NLI). We propose the Polarity Composition Hypothesis, a high-level causal model of this computation. Behavioral evaluations show that language-model predictions largely conform to the proposed hypothesis. Using Distributed Alignment Search, we identify causally effective subspaces aligned with the variables in the polarity-composition hypothesis. These representations transfer across negation forms and partially to a subset of MNLI. We release the dataset and code.

Explore similar work

CardsList
  1. How Language Models Process Negation

    May 4, 2026Zhejian Zhou, Tianyi Zhou, Robin Jia +1LLM InterpretabilityLanguage Modeling

  2. "Not" is Not Lost in Translation: A Causal Analysis of Negation Representations in Multilingual LLMs

    Oct 23, 2026Giang Son Nguyen, Wray Buntine, Dung D. Le +1Cross-Lingual Consistency in Language ModelsMultilingual Language Models