Fine-tuning multilingual models for low-resource dialect translation frequently encounters a “plausibility over faithfulness” dilemma, resulting in severe semantic drift on dialect-specific tokens. We term this phenomenon the “Probability Trap,” where models prioritize statistical fluency over semantic fidelity. To address this, we propose MVS-Rank (Multi-View Scoring Reranking), a generate-then-rerank framework that decouples evaluation from generation. Our method assesses translation candidates through three complementary perspectives: (1) Source-Side Faithfulness via a Reverse Translation Model to anchor semantic fidelity; (2) Local Fluency using Masked Language Models to ensure syntactic precision; and (3) Global Fluency leveraging Large Language Models to capture discourse coherence. Extensive experiments on Cantonese-Mandarin benchmarks demonstrate that MVS-Rank achieves state-of-the-art performance, significantly outperforming strong fine-tuning baselines by effectively rectifying hallucinations while maintaining high fluency.

Existing hard-label text attacks often rely on inefficient “outside-in” strategies that traverse vast search spaces. We propose PivotAttack, a query-efficient “inside-out” framework. It employs a Multi-Armed Bandit algorithm to identify Pivot Sets—combinatorial token groups acting as prediction anchors—and strategically perturbs them to induce label flips. This approach captures inter-word dependencies and minimizes query costs. Extensive experiments across traditional models and Large Language Models demonstrate that PivotAttack consistently outperforms state-of-the-art baselines in both Attack Success Rate and query efficiency.

Word sense disambiguation (WSD) is typically studied in synchronic settings, where the mapping from contextualized occurrences to sense labels is assumed to be temporally stable. This assumption may not hold for historical texts, in which meanings may emerge, shift, or disappear over time. We study diachronic WSD for Ancient Chinese, a challenging scenario involving long-term semantic evolution and substantial historical variation. We introduce Ancient Chinese Diachronic Word Sense Disambiguation Benchmark (AC-DiaWSD), a large-scale WSD dataset constructed from the Twenty-Four Histories, containing about 24K manually annotated instances across 12 dynasties from the Western Han to the Qing. We further propose DynaSense, a model-agnostic framework that enhances pre-trained language models with contrastive sense learning and dynasty-conditioned representation adaptation. Experiments with multiple PLMs and LLMs show that existing models struggle with diachronic WSD, while DynaSense consistently improves accuracy and F1 across backbones.

This paper presents Temporal-aware Soft Prompt Tuning (TASPT), a novel approach for automatic text dating. Unlike existing methods, which often overlook the evolution of word meanings in texts spanning long periods, TASPT incorporates the unique characteristics of historical texts. It introduces a temporal-aware text representation that dynamically captures both semantic variance and invariance. This representation is combined with a soft prompt, enabling efficient parameter tuning for automatic text dating. Experiments show that TASPT outperforms all existing methods on two diachronic datasets: the Twenty-Four Histories and the Royal Society Corpus.