Behavioral and Representational Evidence of Binomial Ordering Preferences in Large Language Models

2026-06-19

AI Summary

This study investigates whether large language models can accurately capture the gradient nature of binomial word order preferences in human language. Formalizing the problem as a distribution alignment task, we construct a dataset of 600 binomial pairs across eight languages and, for the first time, combine behavioral evaluation with sparse representation probing to systematically assess the alignment between model outputs and empirical corpus distributions. Our findings reveal that while models reproduce dominant word orders, they struggle to precisely match the strength of human preferences. We further show that relevant signals are partially encoded in middle-to-late layer representations and can be effectively modulated via probe directions to steer generation distributions. This work highlights limitations in models’ ability to capture fine-grained statistical nuances of language and demonstrates an interpretable method for intervening in their preference behavior.

Abstract

Large language models (LLMs) can readily reproduce conventional expressions, yet their ability to model gradient frequency distributions remains underexplored. We investigate this using linguistic binomials, such as men and women, where both word permutations are grammatically valid but exhibit distinct, cross-linguistic variations in conventionality. We formalize binomial ordering as a distributional alignment problem, and construct a multilingual dataset of 600 binomial pairs across 8 languages. With categorical and distributional metrics, we measure and compare the corpus-derived preferences with model-induced ordering probabilities of 6 open-weight LLMs. While models often behaviorally recover the dominant corpus-preferred order, particularly for strongly conventionalized pairs, they align less well with the exact corpus preference distributions. This suggests that apparent directional order overstates how faithfully LLMs capture the statistical nuances of language use. Sparse probing verifies that the concept of preference strength is partially encoded among middle-to-late layers, and steering along probe-derived directions alters model-induced ordering distributions, demonstrating that the statistical behavioral preference of LLMs can be mechanistically measured and manipulated via internal representations.

Problem

• binomial ordering
• large language models
• distributional alignment
• frequency distribution
• linguistic conventionality

Innovation

• distributional alignment
• binomial ordering
• sparse probing
• preference strength
• representation steering