Skip to content
Preprint

Inference-Time Steering for Cross-Lingual Factual Consistency in LLMs

Jul 2026 · 0 citations · 24 references
Computer Science

TL;DR

It is suggested that cross-lingual inconsistency is at least partly a selection problem, and that simple contextual interventions may outperform more invasive methods for robust, transferable alignment.

Abstract

Although Large Language Models (LLMs) demonstrate remarkable multilingual fluency, their internal knowledge representations remain disproportionately biased toward high-resource languages. This leads to cross-lingual factual inconsistency, where they shift their empirical answer distributions based solely on the prompt language. We investigate whether these biases can be mitigated at inference time, forcing an English-prompted model to answer as if it were queried in target languages (German, Spanish, Bulgarian), and evaluate four intervention strategies: zero-shot contextual steering (persona prompting), internal representation manipulation via Contrastive Activation Addition (CAA), and lightweight weight modification via Direct Preference Optimization (DPO) trained on benchmark-derived factual data as well as conceptual generalization data. To assess alignment, we curate a multilingual factual dataset alongside a novel generalization benchmark comprising culturally rooted queries to determine whether factual interventions transfer to broader target-centric preferences. Experiments on Gemma 3 12B Instruct reveal persona prompting to be the strongest overall intervention, balancing efficacy, safety, and out-of-domain generalization. While CAA yields sharp inconsistency benchmark shifts, it is configuration-sensitive and risks knowledge degradation. DPO-based adapters offer permanent, yet narrower and less transferable gains. These findings suggest that cross-lingual inconsistency is at least partly a selection problem, and that simple contextual interventions may outperform more invasive methods for robust, transferable alignment.

View source

Similar papers

Preprint Jul 2026

Knowledgeless Language Models: Suppressing Parametric Recall for Evidence-Grounded Language Modeling

The results demonstrate that suppressing entity-linked supervision during pretraining induces a shift in epistemic behavior: KLLMs rely less on parametric knowledge and more on external evidence, leading to improved reliability under realistic conditions, suggesting that pretraining-time control over knowledge acquisition can complement retrieval-augmented and tool-based systems by providing a more evidence-sensitive base model.

Roi Cohen, Yvan Carr'e, Nick Lechtenborger et al. · 0 citations
Preprint Aug 2026

Cross-lingual Representation Learning via Centroid Intervention Fusion

Centroid Intervention Fusion is proposed, a projection fusion framework that consolidates multiple multilingual intervention projections into a single language-shared operator and outperforms the strongest prior pairwise intervention baseline by up to +3.3% across four model backbones.

Wei Sun, Marie-Francine Moens · 0 citations
Open access Aug 2026

Towards Trustworthy Large Language Models

An integrated conceptual frame-work that couples attention- and perturbation-based explainability with lightweight hallucination-detection signals and token-efficient inference strategies is presented, and a set of cross-cutting consistency metrics are instrumented with a set of cross-cutting consistency metrics.

Sakshi Parate, Shreyans Sanyal · 0 citations
Preprint Aug 2026

When the API Speaks the Wrong Language: Revisiting Post-Training for Multilingual Tool Use

It is found that, in this benchmark, supervised fine-tuning (SFT) provides a strong baseline, substantially improving argument language consistency and end-to-end function call accuracy and, under consistent model selection, SFT achieves performance comparable to, and sometimes exceeding more complex reinforcement learning (RL) approaches.

Siddharth Chauhan, Thomas Butler, Abhishek Singhania et al. · 0 citations

Probing cross-lingual differences in how LLMs represent Natural Language Inference

Cross-lingual probe transfer and direct comparison of probe weights show that NLI representations are most alignable in the middle layers: probes transfer best there, and probes trained independently on different languages converge to similar weight vectors, peaking mid-network.

Nicolas Ramos Fernandez · 0 citations
Preprint Jul 2026

Rationale-Guided Knowledge Distillation for Cross-Lingual Stance Detection

Stance detection aims to identify whether a text expresses a favorable or opposing attitude toward a given target, and serves as an important task for various downstream applications. Although existing studies have achieved strong performance in monolingual settings, especially in English, many low-resource languages such as Catalan still lack sufficient annotated data for training effective models. Cross-lingual stance detection alleviates this problem by transferring stance knowledge from resource-rich languages to low-resource languages. However, most existing methods mainly rely on semantic alignment between texts and targets, while ignoring the reasoning process required for reliable stance inference. Although Large Language Models provide strong reasoning ability, their high computational cost and inference latency limit practical deployment. To address these limitations, we propose a rationale-guided knowledge distillation framework for cross-lingual stance detection. Specifically, we use Chain-of-Thought prompting to guide Large Language Models in generating informative rationales, and distill the resulting reasoning knowledge into a compact student model. We further design a dual-path distillation mechanism to align rationale-enhanced and rationale-free representations, together with their prediction distributions. In addition, two contrastive learning strategies are introduced to improve stance discrimination. Experiments on multilingual benchmarks demonstrate that our method consistently outperforms competitive baselines.

Qiuli Zhou, Jingyuan Yao, Shengeng Tang et al. · 0 citations