Beyond the Final Layer: Intermediate Representations for Better Multilingual Calibration in Large Language Models
By: Ej Zhou , Caiqi Zhang , Tiancheng Hu and more
Potential Business Impact:
Makes AI understand other languages better.
Confidence calibration, the alignment of a model's predicted confidence with its actual accuracy, is crucial for the reliable deployment of Large Language Models (LLMs). However, this critical property remains largely under-explored in multilingual contexts. In this work, we conduct the first large-scale, systematic studies of multilingual calibration across six model families and over 100 languages, revealing that non-English languages suffer from systematically worse calibration. To diagnose this, we investigate the model's internal representations and find that the final layer, biased by English-centric training, provides a poor signal for multilingual confidence. In contrast, our layer-wise analysis uncovers a key insight that late-intermediate layers consistently offer a more reliable and better-calibrated signal. Building on this, we introduce a suite of training-free methods, including Language-Aware Confidence Ensemble (LACE), which adaptively selects an optimal ensemble of layers for each specific language. Our study highlights the hidden costs of English-centric alignment and offer a new path toward building more globally equitable and trustworthy LLMs by looking beyond the final layer.
Similar Papers
Calibration Across Layers: Understanding Calibration Evolution in LLMs
Machine Learning (CS)
Makes AI more honest about what it knows.
Cross-Lingual Stability and Bias in Instruction-Tuned Language Models for Humanitarian NLP
Computation and Language
Helps find human rights abuses in any language.
Beyond Accuracy: The Role of Calibration in Self-Improving Large Language Models
Computation and Language
Makes AI more honest about what it knows.