Weixiong Zheng, Peijian Zeng, YiWei Li, Hongyan Wu, Nankai Lin, Junhao Chen, Aimin Yang, Yongmei Zhou. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The performance of multilingual language models (MLLMs) is notably inferior for low-resource languages (LRL) compared to high-resource ones, primarily due to the limited available corpus during the pre-training phase. This inadequacy stems from the under-representation of low-resource language words in the subword vocabularies of MLLMs, leading to their misidentification as unknown or incorrectly concatenated subwords. Previous approaches are based on frequency sorting to select words for augmenting vocabularies. However, these methods overlook the fundamental disparities between model representation distributions and frequency distributions. To address this gap, we introduce a novel Entropy-Consistency Word Selection (ECWS) method, which integrates semantic and frequency metrics for vocabulary augmentation. Our results indicate an improvement in performance, supporting our approach as a viable means to enrich vocabularies inadequately represented in current MLLMs.

Understanding and modeling time remains a key challenge in today's video understanding systems. With language playing a central role in driving powerful generalization, foundational video-language models must inherently capture a sense of temporality. However, Video-text retrieval faces three challenges: Parameter efficiency bottlenecks (traditional prompt learning requires adjusting over 10 % of parameters, leading to a sharp increase in computational costs), limitations of static course strategies (manually defined difficulty thresholds cause pseudo-label biases in highly abstract actions), and insufficient adaptability to heterogeneous architectures (cross-modal hybrid expert frameworks rely on shared weights, restricting the flexible combination of heterogeneous models such as CLIP and ViT). To address these challenges, this paper proposes the Dynamic Curriculum Learning with Adaptive Prompt Mixture of Experts (DynaCLIP) framework. Through the collaborative optimization of the Adaptive Prompt Mixture of Experts (APMoE) module and dynamic curriculum learning, the parameter efficiency and dynamic adaptability are significantly improved. APMoE adopts orthogonal initialization expert pools and cross-modal routing networks. It freezes the pre-trained backbone networks (such as CLIP/BERT) on the premise of fine-tuning only less than 1 % of parameters (static expert pool + dynamic prompt generator), and generates instance-aware text encoder prompts by dynamically selecting Top-k experts and weighted fusion. Here, these instanceaware prompts are implemented as continuous soft prompts in the embedding space, rather than discrete natural language text. Meanwhile, a two-dimensional difficulty quantification system is constructed based on the verb abstraction level (VerbNet three-level classification) and BERT semantic similarity. Combined with the improved binary search strategy to balance high-confidence samples and diversity exploration, a cognitiveinspired course learning mechanism is formed. Experiments show that DynaCLIP achieves 7.4 % R@1 on the TEMPO dataset, 67.5 % accuracy in temporal inference. The code is available at: https://github.com/18162195164/DynaCLIP.

Industrial Anomaly Detection (IAD) is critical for ensuring product quality by identifying defects. Traditional methods such as feature embedding and reconstruction-based approaches require large datasets and struggle with scalability. Existing vision-language models (VLMs) and Multimodal Large Language Models (MLLMs) address some limitations but rely on mask annotations, leading to high implementation costs and false positives. Additionally, industrial datasets like MVTec-AD and VisA suffer from severe class imbalance, with defect samples constituting only 23.8 % and 11.1 % of total data respectively. To address these challenges, we propose a reward function that dynamically prioritizes rare defect patterns during training to handle class imbalance. We also introduce a mask-free reasoning framework using Chain of Thought (CoT) and Group Relative Policy Optimization (GRPO) mechanisms, enabling anomaly detection directly from raw images without annotated masks. This approach generates interpretable step-by-step explanations for defect localization. Our method achieves state-of-the-art performance, outperforming prior approaches by 36% in accuracy on MVTec-AD and 16% on VisA. By eliminating mask dependency and reducing costs while providing explainable outputs, this work advances industrial anomaly detection and supports scalable quality control in manufacturing.

Abstract Curriculum learning (CL) is a training strategy that imitates how humans learn, by gradually introducing more complex samples and information to the model. However, in multi-label emotion classification (MEC) tasks, using a traditional CL approach can result in overfitting on easy samples and lead to biased training. Additionally, the sample difficulty varies as the model trains. To address these challenges, we propose a novel CL framework for MEC tasks called CLF-MEC. Unlike traditional approaches that assess difficulty at the sample level, we utilize category-level assessment to determine the difficulty level of samples. As the model identifies a category well, the score for that category’s samples is reduced, ensuring dynamic changes in the sample difficulty are accounted for. Our CL framework employs two training modes, namely “learning” and “tackling.” These two processes are trained alternatively to imitate the “learning-tackling” process in human learning. This ensures that samples from hard-to-learn categories receive more attention. During the “tackling” process, our method transforms the task of dealing with hard samples into an “easy” learning task by utilizing contrastive learning to enhance the semantic representation of those hard samples. Experimental results demonstrate that our CLF-MEC framework has achieved significant improvements in MEC.