arXiv subjects

Jingshen Zhang

Publications and source records attributed to Jingshen Zhang.

11 recordsLinked to original sources

arXiv2026

Aligned but Flattened: Analyzing the Trade-off between Cultural Alignment and Diversity in LLMs

Cultural fine-tuning has become the de facto paradigm for building culture-aware large language models (LLMs), yet existing optimization exclusively for alignment scores provides an incomplete portrait of cultural fidelity by systematically obscuring inherent cultural diversity. This unidimensional evaluation lens prompts a fundamental question: do models genuinely perceive distinct cultural nuances, or do they merely memorize dominant cultural values? To address this, we propose a synergistic evaluation framework that jointly formalizes cultural alignment and diversity. Through extensive benchmarking of six mainstream LLMs on the World Values Survey, this framework uncovers a systematic and critical trade-off: the pursuit of cultural alignment consistently incurs an acute expense of diversity, leading to severe "cultural flattening." Investigating this behavioral shift, we demonstrate that these superficial alignment gains stem from models artificially anchoring to dominant majorities, converging onto a monolithic response pattern that wipes out the heterogeneous distributions inherent to human groups. Crucially, our mechanistic analysis suggests that this diversity collapse is not merely a behavioral anomaly but more likely a structural consequence of the low-rank bias inherent in neural network optimization. Therefore, our findings expose the limitations of current post-training paradigms and call for a shift toward alignment objectives that preserve cross-cultural pluralism.

cs.SI↗

arXiv2026

Beyond Alignment: Value Diversity as a Collective Property in Multicultural Agent Systems

Multicultural multi-agent systems are increasingly deployed in globally diverse settings, where different agents are grounded in different cultural backgrounds. Existing cultural evaluation focuses on value alignment: how closely a single agent matches a target culture. Yet alignment is a per-agent property and cannot reveal whether a system, taken as a whole, preserves the cultural plurality it is meant to represent. We propose value diversity as a system-level evaluation axis for multicultural agent systems, defined through the dissimilarity between culturally conditioned agents' responses on a shared value survey. Using the World Values Survey, we evaluate 19 cultures and 18 backbone models across a wide range of system configurations. We find that diversity is largely uncorrelated with alignment, indicating that the two capture complementary system properties, and that current multicultural agent systems fall substantially below human societies in value diversity. Mixed-backbone systems narrow this gap but do not close it, and the gap persists across culture compositions and agent scales. Social interaction further erodes diversity by driving agents toward consensus, and a participatory budgeting case study shows that this homogenization narrows the breadth of collective decision-making. Together, our results establish value diversity as a distinct evaluation axis for multicultural multi-agent systems and reveal a persistent homogenization tendency in current LLM-based societies. Our code and data are publicly available at https://github.com/iNLP-Lab/MultiAgent-Diversity.

cs.CL↗

arXiv2026

Understanding the Self-Reflection Mechanisms of LLMs through Biased Attitude Associations

While the emergent self-reflection capabilities of Large Language Models (LLMs) offer a promising paradigm for autonomous bias mitigation, their internal mechanics remain unclear, raising concerns regarding potential bias entrenchment. Under the premise that social bias is intrinsically encoded as valence inclinations, where the exacerbation of bias scales with sharper valence fluctuations across social groups, this paper proposes ReBias-Lens, a probing framework designed to interpret how self-reflection reconfigures these biased attitude associations through the lens of valence projection within intersectional contexts. Central to ReBias-Lens is the metric of Valence Fluctuation (VF) comprising two variants: Global-VF, which captures macroscopic valence encoding trends, and Local-VF, which scrutinizes microscopic distinctiveness across specific social categories. Deploying ReBias-Lens to evaluate four LLMs across twelve social categories reveals that overall valence fluctuations undergo a distinct layer-wise smoothing, characterized by a significant hierarchical representation divergence as the layers deepen, which ultimately manifests as a widespread mitigation of bias at the behavioral level. In stark contrast to this macro-level reduction, this reflection mechanism is not universally corrective, instead exhibiting a stubborn, category-specific selectivity that regularly locks in and perversely amplifies localized biases. Warning: this paper contains examples with biased content.

cs.SI↗

arXiv2026

Modeling Implicit Conflict Monitoring Mechanisms against Stereotypes in LLMs

In this paper, we study an emergent self-debiasing mechanisms against stereotypical content in Large Language Models (LLMs). Unlike traditional safety mechanisms that are primarily triggered by explicit input-level stimuli, self-debiasing mechanisms can involve generation-time intrinsic correction that are not directly reducible to surface-level prompt. Motivated by conflict-monitoring and response-inhibition accounts in cognitive neuroscience, we propose COCO, a contrastive causal method designed to identify COCO neurons that exhibit high intra-\underline{CO}nsistency yet sharp inter-\underline{CO}ntrast across antithetical generative responses, such as stereotypical versus unbiased outputs. Ablation studies reveal that deactivating COCO neurons leads to a catastrophic collapse of the model's fairness; over 90\% of outputs revert to biased content, far exceeding the bias levels induced by explicit adversarial jailbreak attacks. Observing that simple weight amplification of COCO neurons yields only marginal gains, we propose two training-free, lightweight editing strategies: Local Enhancement (LE-COCO) and Networked Enhancement (NE-COCO). Comprehensive evaluations show that our methods bolster robustness against adversarial jailbreaks and achieve strong performance on open-ended safety benchmarks, while preserving foundational generative proficiency. While this study primarily addresses social stereotypes, the COCO mechanism holds significant potential for diverse domains like hallucination detection, offering valuable insights toward the development of self-evolving AI agents.

cs.SI↗

arXiv2026

Let's Simplify Step by Step: Guiding LLM Towards Multilingual Unsupervised Proficiency-Controlled Sentence Simplification

Large language models demonstrate limited capability in proficiency-controlled sentence simplification, particularly when simplifying across large readability levels. We propose a framework that decomposes complex simplifications into manageable steps through dynamic path planning, semantic-aware exemplar selection, and chain-of-thought generation with conversation history for coherent reasoning. Evaluation on five languages across two benchmarks shows our approach improves simplification effectiveness while reducing computational steps by 22-42%. Human evaluation confirms the fundamental trade-off between simplification effectiveness and meaning preservation. Notably, even human annotators struggle to agree on semantic preservation judgments, highlighting the inherent complexity of this task. Our work shows that while step-by-step simplification improves control, preserving semantic fidelity during extensive simplification remains an open challenge.

cs.CL↗

arXiv2025

DualReward: A Dynamic Reinforcement Learning Framework for Cloze Tests Distractor Generation

This paper introduces DualReward, a novel reinforcement learning framework for automatic distractor generation in cloze tests. Unlike conventional approaches that rely primarily on supervised learning or static generative models, our method employs a dual reward structure with adaptive scaling that differentiates between human-created gold standard distractors and model-generated candidates. The framework dynamically adjusts reward signal intensity based on model performance and confidence. We evaluate our approach on both passage-level (CLOTH-F) and sentence-level (MCQ) cloze test datasets, demonstrating consistent improvements over state-of-the-art baselines. Experimental results show that our adaptive reward scaling mechanism provides modest but consistent benefits on homogeneous datasets (CLOTH-F) and more substantial improvements (3.48-3.86% in P@1) on diverse, cross-domain data (MCQ), suggesting its particular effectiveness for handling varied question types and domains. Our work offers a flexible framework that effectively balances learning from reliable human examples while exploring novel, high-quality distractors for automated test generation.

cs.CL↗

arXiv2024

Label Confidence Weighted Learning for Target-level Sentence Simplification

Multi-level sentence simplification generates simplified sentences with varying language proficiency levels. We propose Label Confidence Weighted Learning (LCWL), a novel approach that incorporates a label confidence weighting scheme in the training loss of the encoder-decoder model, setting it apart from existing confidence-weighting methods primarily designed for classification. Experimentation on English grade-level simplification dataset shows that LCWL outperforms state-of-the-art unsupervised baselines. Fine-tuning the LCWL model on in-domain data and combining with Symmetric Cross Entropy (SCE) consistently delivers better simplifications compared to strong supervised methods. Our results highlight the effectiveness of label confidence weighting techniques for text simplification tasks with encoder-decoder architectures.

cs.CL↗

arXiv2024

System Report for CCL24-Eval Task 7: Multi-Error Modeling and Fluency-Targeted Pre-training for Chinese Essay Evaluation

This system report presents our approaches and results for the Chinese Essay Fluency Evaluation (CEFE) task at CCL-2024. For Track 1, we optimized predictions for challenging fine-grained error types using binary classification models and trained coarse-grained models on the Chinese Learner 4W corpus. In Track 2, we enhanced performance by constructing a pseudo-dataset with multiple error types per sentence. For Track 3, where we achieved first place, we generated fluency-rated pseudo-data via back-translation for pre-training and used an NSP-based strategy with Symmetric Cross Entropy loss to capture context and mitigate long dependencies. Our methods effectively address key challenges in Chinese Essay Fluency Evaluation.

cs.CL↗

arXiv2024

Cross-Lingual Word Alignment for ASEAN Languages with Contrastive Learning

Cross-lingual word alignment plays a crucial role in various natural language processing tasks, particularly for low-resource languages. Recent study proposes a BiLSTM-based encoder-decoder model that outperforms pre-trained language models in low-resource settings. However, their model only considers the similarity of word embedding spaces and does not explicitly model the differences between word embeddings. To address this limitation, we propose incorporating contrastive learning into the BiLSTM-based encoder-decoder framework. Our approach introduces a multi-view negative sampling strategy to learn the differences between word pairs in the shared cross-lingual embedding space. We evaluate our model on five bilingual aligned datasets spanning four ASEAN languages: Lao, Vietnamese, Thai, and Indonesian. Experimental results demonstrate that integrating contrastive learning consistently improves word alignment accuracy across all datasets, confirming the effectiveness of the proposed method in low-resource scenarios. We will release our data set and code to support future research on ASEAN or more low-resource word alignment.

cs.CL↗

arXiv2024

Readability-guided Idiom-aware Sentence Simplification (RISS) for Chinese

Chinese sentence simplification faces challenges due to the lack of large-scale labeled parallel corpora and the prevalence of idioms. To address these challenges, we propose Readability-guided Idiom-aware Sentence Simplification (RISS), a novel framework that combines data augmentation techniques with lexcial simplification. RISS introduces two key components: (1) Readability-guided Paraphrase Selection (RPS), a method for mining high-quality sentence pairs, and (2) Idiom-aware Simplification (IAS), a model that enhances the comprehension and simplification of idiomatic expressions. By integrating RPS and IAS using multi-stage and multi-task learning strategies, RISS outperforms previous state-of-the-art methods on two Chinese sentence simplification datasets. Furthermore, RISS achieves additional improvements when fine-tuned on a small labeled dataset. Our approach demonstrates the potential for more effective and accessible Chinese text simplification.

cs.CL↗

arXiv2024

Controlling Cloze-test Question Item Difficulty with PLM-based Surrogate Models for IRT Assessment

Item difficulty plays a crucial role in adaptive testing. However, few works have focused on generating questions of varying difficulty levels, especially for multiple-choice (MC) cloze tests. We propose training pre-trained language models (PLMs) as surrogate models to enable item response theory (IRT) assessment, avoiding the need for human test subjects. We also propose two strategies to control the difficulty levels of both the gaps and the distractors using ranking rules to reduce invalid distractors. Experimentation on a benchmark dataset demonstrates that our proposed framework and methods can effectively control and evaluate the difficulty levels of MC cloze tests.

cs.CL↗