# Choosing the Right Language Mode at Inference Time for Multilingual Reliability

arXiv: https://arxiv.org/abs/2609.04653 · HTML: https://mlsanity.com/paper/2609.04653/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 29 references that are in the map.
- First posted on arXiv 4 Sep 2026.

## Builds on (most cited first, 22 of 22)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants](https://mlsanity.com/paper/2308.16884/) (2023)
- [Do Llamas Work in English? On the Latent Language of Multilingual Transformers](https://mlsanity.com/paper/2402.10588/) (2024)
- [A Survey of Confidence Estimation and Calibration in Large Language Models](https://mlsanity.com/paper/2311.08298/) (2023)
- [Not All Languages Are Created Equal in LLMs: Improving Multilingual Capability by Cross-Lingual-Thought Prompting](https://mlsanity.com/paper/2305.07004/) (2023)
- [AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions](https://mlsanity.com/paper/2506.09038/) (2025)
- [Cross-lingual Prompting: Improving Zero-shot Chain-of-Thought Reasoning across Languages](https://mlsanity.com/paper/2310.14799/) (2023)
- [On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons](https://mlsanity.com/paper/2404.02431/) (2024)
- [Uncertainty-Based Abstention in LLMs Improves Safety and Reduces Hallucinations](https://mlsanity.com/paper/2404.10960/) (2024)
- [Question Translation Training for Better Multilingual Reasoning](https://mlsanity.com/paper/2401.07817/) (2024)
- [Layer Swapping for Zero-Shot Cross-Lingual Transfer in Large Language Models](https://mlsanity.com/paper/2410.01335/) (2024)
- [Language-Specific Latent Process Hinders Cross-Lingual Performance](https://mlsanity.com/paper/2505.13141/) (2025)
- [On the Calibration of Massively Multilingual Language Models](https://mlsanity.com/paper/2210.12265/) (2022)
- [Teaching LLMs to Abstain across Languages via Multilingual Feedback](https://mlsanity.com/paper/2406.15948/) (2024)
- [Beyond the Final Layer: Intermediate Representations for Better Multilingual Calibration in Large Language Models](https://mlsanity.com/paper/2510.03136/) (2025)
- [Evaluating Multilingual Long-Context Models for Retrieval and Reasoning](https://mlsanity.com/paper/2409.18006/) (2024)
- [On the Calibration of Multilingual Question Answering LLMs](https://mlsanity.com/paper/2311.08669/) (2023)
- [Beyond Data Quantity: Key Factors Driving Performance in Multilingual Language Models](https://mlsanity.com/paper/2412.12500/) (2024)
- [Cross-Lingual Activation Steering for Multilingual Language Models](https://mlsanity.com/paper/2601.16390/) (2026)
- [CAPO: Confidence Aware Preference Optimization Learning for Multilingual Preferences](https://mlsanity.com/paper/2511.07691/) (2025)

## Abstract (arXiv)

Multilingual large language models often struggle to reason in low- to mid-resource languages. Prior work has shown that translation can improve multilingual reasoning by helping models access stronger English-centric representations. This raises a central question: How much translation is needed for multilingual large language models to reason reliably, and when does more translation instead trigger interference and overconfidence? Using LLaMA and Qwen models, we run extensive experiments varying text scope and language mode (target-only, English-only, bilingual) to evaluate both accuracy and reliability. Our results reveal a clear trade-off: English context often improve understanding and recover errors caused by non-English comprehension, yet adding redundant bilingual context intensifies interference. We address this trade-off with Reliability-Aware Adaptive Inference (RAAI), a training-free test-time framework that (i) performs Expected Calibration Error (ECE)-aware routing and prompt fusion, and (ii) uses a mid-layer Risk Index (RI) to gate sequential reasoning, allocating compute only when it is likely to help and suppressing harmful bilingual redundancy. Across two model families, RAAI enhances accuracy by 25-37.7% on low-resource languages and lowers calibration error by approximately 3-6%, with the most pronounced benefits in the lowest-resource language tiers.