Skeleton-based reasoning prompting is a promising training-free approach for structuring LLM reasoning, but prior work largely assumes an English-centric setting. We propose the Language-Aware Skeleton Exploration Framework () to study skeleton-language choice in multilingual mathematical reasoning. Across math benchmarks, model scales, and languages, we show that English skeletons yield a small positive tendency on average, most visible for smaller models and low-resource languages. However, few language-level gains remain significant after correction, and English is not universally optimal. Combining greedy decoding, multi-rollout evaluation, translation ablation, and cross-benchmark validation, we further find three patterns of skeleton-language effects: directionally consistent, evaluation- and benchmark-dependent, and asymmetric negative. These effects cannot be fully explained by generation quality alone. Overall, skeleton language is a context-dependent design variable that requires multi-level exploration. All resources are released[<https://github.com/lhsstn/LASEF>].

Recent advancements in Large Language Models (LLMs) have shown remarkable capabilities in multilingual contexts. However, previous studies suggest that multilingual LLMs primarily rely on the rich semantic knowledge of high-resource (i.e., English) representations to conduct tasks across languages. To mitigate the performance gap arising from resource imbalances, we propose Steering via Language-Specific Neurons, a training-free method that aligns non-English representations with the English manifold intervening on language-specific neuron activations via linear interpolation toward English activation prototypes. Experimental results show that the proposed method generally improves downstream task performance with minimal computational cost, yielding performance gains for low-resource scenarios where limited data often constrains knowledge coverage. Our findings suggest that the fundamental limitations of low-resource languages can be effectively alleviated by leveraging the robust representation space of high-resource languages, offering a promising direction for more balanced multilingual LLMs.

Synthetic data is increasingly used to train large language models (LLMs), yet its security implications remain poorly understood. Prior work on subliminal learning suggests that models can inherit behavioral traits from seemingly unrelated training data. In this work, we investigate whether such mechanisms can be exploited to inject targeted social biases into aligned models through semantically benign synthetic data. We construct a pipeline in which a misaligned teacher model generates filtered synthetic datasets across domains such as creative writing and code generation, which are then used to fine-tune aligned student models. Our experiments show that benign-looking synthetic data can act as a covert channel for transmitting targeted biases while largely preserving the student model’s general task capabilities. These results reveal a previously underexplored security risk in synthetic data–driven LLM training pipelines and highlight the need for improved safeguards. As one possible step toward this goal, we suggest that log-linearity-based scoring may provide a useful signal for screening seemingly benign synthetic data.

Multilingual large language models (mLLMs) contain a small fraction of feed-forward neurons that are sensitive to particular languages, commonly termed language-specific neurons. Existing work measures language specificity using the entropy of each neuron’s language-wise probabilities of being active, where a neuron is considered active when its activation value is positive. However, this approach may not fully capture the multilingual nature of mLLMs, where language representations are distributional and mutually related. We propose Distribution-aware Language Neuron selection, which leverages pairwise relationships between per-language activation distributions over the full activation range, including negative values. Specifically, we quantify each neuron’s language specificity by clustering languages using pairwise overlap coefficients between their activation distributions. Across two mLLMs and two held-out corpora, our identifier more effectively isolates language-specific causal effects, yielding up to 4.9 times higher on-target language damage per neuron while preserving off-target language performance.