Large Language Models (LLMs) frequently exhibit strong translation abilities, even without task-specific fine-tuning. However, the internal mechanisms governing this innate capability remain largely opaque. To demystify this process, we leverage Sparse Autoencoders (SAEs) and introduce a novel framework for identifying task-specific features. Our method first recalls features that are frequently co-activated on translation inputs and then filters them for functional coherence using a PCA-based consistency metric. This framework successfully isolates a small set of "translation initiation" features. Causal interventions demonstrate that amplifying these features steers the model towards correct translation, while ablating them induces hallucinations and off-task outputs, confirming they represent a core component of the model's innate translation competency. Moving from analysis to application, we leverage this mechanistic insight to propose a new data selection strategy for efficient fine-tuning. Specifically, we prioritize training on "mechanistically hard" samples—those that fail to naturally activate the translation initiation features. Experiments show this approach significantly improves data efficiency and suppresses hallucinations. Furthermore, we find these mechanisms are transferable to larger models of the same family. Our work not only decodes a core component of the translation mechanism in LLMs but also provides a blueprint for using internal model mechanism to create more robust and efficient models.

Large language models (LLMs), have shown strong potential in scientific discovery, yet existing methods still face substantial challenges in the design of research workflows and multi-role collaboration mechanisms.To mitigate these issues, we propose EvoSci, a multi-agent scientific collaboration framework, which integrates bio-inspired evolution with knowledge graph modeling.To iteratively generate, evaluate, and refine research ideas, EvoSci incorporates multiple role-based agents, including mentor, researcher, and reviewer.By combining collaborative reasoning, shared memory, and evolutionary feedback, EvoSci significantly enhances the coherence and creativity of scientific exploration.Experiments on real-world research topics demonstrate that EvoSci significantly outperforms strong baselines in LLM-based structured peer-review and comparative ranking evaluations, achieving the highest overall peer-review score (ICLR 4.90) and top ranking (Top-10 = 54).These results suggest its superiority in both scientific idea generation and continuous discovery.

Lei Yang, Leiyu Pan, Bojian Xiong, Renren Jin, Shaowei Zhang, Yue Chen, Ling Shi, Jiang Zhou, Junru Wu, Zhen Wang, Jianxiang Peng, Juesi Xiao, Tianyu Dong, Zhuowen Han, Zhuo Chen, Yuqi Ren, Deyi Xiong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large Language Models (LLMs) are increasingly deployed in contexts requiring complex moral reasoning and value trade-offs.However, existing evaluations typically rely on item-level behavioral metrics, which fail to capture how models structurally prioritize competing values as a cohesive system.To address this, we propose a symmetric human-LLM evaluation framework, grounded in Q methodology, to measure value-structure alignment.Under our protocol, humans and models sort an identical 140-item moral statement set into a shared nine-column forced distribution; for LLMs, we elicit strict rankings and deterministically map them to Q-sort buckets.Using a human reference sample (N = 35), we establish a stable three-factor reference geometry specific to this instrument and sample.We evaluate 12 LLMs across four model families via 240 replicated Q-sorts at two temperature settings, quantifying structural alignment via Procrustes similarity (ϕ) and RSA-based Spearman correlation (ρ).Our results reveal significant crossfamily heterogeneity, model-specific sensitivity to generation stochasticity and localized misalignment, which demonstrate that favorable global scores can obscure underlying regional distortions.While rank-and bucket-based analyses remain highly consistent, prompt phrasing introduces notable variance.Ultimately, assessing value-structure alignment provides a crucial structural complement to traditional itemwise moral benchmarks.

Current Large Language Models (LLMs) typically rely on coarse-grained national labels for pluralistic value alignment.However, such macro-level supervision often obscures intra-country value heterogeneity, yielding a loose alignment.We argue that resolving this limitation requires shifting from national labels to multi-dimensional demographic constraints, which can identify groups with predictable, high-consensus value preference.To this end, we propose DVMap (High-Consensus Demographic-Value Mapping), a framework for fine-grained pluralistic value alignment.In this framework, we first present a demographic archetype extraction strategy to construct a high-quality value alignment corpus of 56,152 samples from the World Values Survey (WVS) by strictly retaining respondents with consistent value preferences under identical demographics.Over this corpus, we introduce a Structured Chain-of-Thought (CoT) mechanism that explicitly guides LLMs to reason about demographic-value correlations.Subsequently, we employ Group Relative Policy Optimization (GRPO) to achieve adaptive anchoring of value distributions.To rigorously evaluate generalization, we further establish a triple-generalization benchmark (spanning cross-demographic, cross-country, and crossvalue) comprising 21,553 samples.Experimental results demonstrate that DVMap effectively learns the manifold mapping from demographics to values, exhibiting strong generalization and robustness.On cross-demographic tests, Qwen3-8B-DVMap achieves 48.6% accuracy, surpassing the advanced open-source LLM DeepSeek-v3.2(45.1%).The source code and dataset are available at https://github. com/EnlightenedAI/DVMap.

Micro-holes with small diameters and large aspect ratios require endoscopic probes for deep morphological inspection. Accurate identification of the micro-hole center is essential for precise probe alignment. However, current industrial production lines rely on manual alignment, a time-consuming and labor-intensive process with limited accuracy. Moreover, existing Hough transform-based circle detection methods exhibit poor robustness, often failing in real-world environments, which leads to alignment errors. To address these challenges, this paper proposes a high-precision micro-hole circle detection method that integrates a lightweight focusable network (LF-Net) with a resolution-aware Hough transform (A-RAHT). The method first generates precise micro-hole masks and then efficiently detects circles. Specifically, LF-Net employs a global grouped linear embedding structure and a dynamic channel sparsity mechanism to reduce computational complexity while effectively modeling long-range dependencies, capturing more representative high-level features of micro-hole circles. Then, an adaptive Sobel operator combined with a multiscale receptive field focusable module is introduced to accurately extract circle masks. Finally, A-RAHT partitions the image into multiple regions and applies multi-threaded acceleration alongside a coarse-to-fine resolution-aware Hough transform, enabling accurate and efficient circle detection. Experimental results demonstrate that the proposed method achieves state-of-the-art performance in detecting micro-hole circles in real industrial environments, significantly enhancing automation and precision. Furthermore, additional experiments across various scenarios validate its robustness and generalization capabilities, making it well-suited for high-precision and real-time circle detection in diverse practical applications.

Large Language Models (LLMs) have demonstrated impressive performance across diverse natural language processing (NLP) and reasoning tasks, yet the influence of psychological factors, such as personality traits, on their capabilities remains underexplored. Drawing on the Big Five personality framework, we investigate how personality configurations affect LLM performance across five task categories: interdisciplinary expert knowledge, safety and harmfulness detection, code generation, mathematical reasoning, and scientific problem solving. We control personality traits using two approaches: prompt-based induction with expert-crafted prompts and low-rank adaptation (LoRA) fine-tuning on a newly constructed dataset of 20,000 personality-conditioned instructions. Experiments on seven LLMs reveal systematic, task-dependent effects of personality. High Neuroticism consistently degrades robustness and generalization, while high Conscientiousness improves stability, particularly in safety-critical contexts. Larger models show stronger resilience under personality perturbations, and prompt-based control achieves a better balance between trait alignment and task performance than LoRA fine-tuning. These findings highlight the trade-offs between controllability, stability, and generalization in personality-aware LLMs.

The behavioral complexity of large language models (LLMs) has sparked growing interest in whether these models mirror human psychological traits. While prior work has explored the presence of personality in LLMs, most studies remain superficial, focus on linguistic style or isolated benchmarks, without probing deeper cognitive alignment with humans. In this paper, we present a comprehensive investigation into whether and how personality traits influence LLMs' behavior. Grounded in the Big Five personality traits, we introduce personalityconditioned prompts and evaluate their effects across both closed (e.g., reasoning, coding) and open-ended (e.g., writing) tasks. Our analysis spans task performance, linguistic style variation, alignment with human personality-ability correlations, and changes in internal reasoning structure. Experimental results reveal that: (i) personality traits affect LLM performance across all tasks, but only influence linguistic style in open-ended generation; (ii) personality-ability correlations in LLMs are broadly consistent with patterns observed in human psychology; and (iii) personality traits alter the behavior of LLMs, leading to different ways of reasoning across traits.

Large Language Models (LLMs) have demonstrated remarkable abilities in text comprehension and logical reasoning, indicating that the text representations learned by LLMs can facilitate their language processing capabilities. In neuroscience, brain cognitive processing signals are typically utilized to study human language processing. Therefore, it is natural to ask how well the text embeddings from LLMs align with the brain cognitive processing signals, and how training strategies affect the LLM-brain alignment? In this paper, we employ Representational Similarity Analysis (RSA) to measure the alignment between 23 mainstream LLMs and fMRI signals of the brain to evaluate how effectively LLMs simulate cognitive language processing. We empirically investigate the impact of various factors (e.g., pre-training data size, model scaling, alignment training, and prompts) on such LLM-brain alignment. Experimental results indicate that pre-training data size and model scaling are positively correlated with LLM-brain similarity, and alignment training can significantly improve LLM-brain similarity. Explicit prompts contribute to the consistency of LLMs with brain cognitive language processing, while nonsensical noisy prompts may attenuate such alignment. Additionally, the performance of a wide range of LLM evaluations (e.g., MMLU, Chatbot Arena) is highly correlated with the LLM-brain similarity.

Partial point cloud registration is an essential and fundamental component of generating complete 3D shapes, aiming at converting partial scans into a unified coordinate system. However, existing point cloud registration methods suffer from inadequately rich local feature extraction and feature interaction. In addition, these methods still face challenges in modeling the global contextual information of point clouds sufficiently, which limits the improvement in registration effectiveness, especially for partial-to-partial point cloud registration with high noise. To overcome these issues, this paper proposes an enhanced PointNet and assignable weights transformer network (LFA-Net) for partial point cloud registration. The model achieves coarse-to-fine point cloud registration through three core modules. First, the Sufficient Local Feature Extraction Module (LFM) is constructed to extract various local feature information. Then, the Adequate Feature Aggregation Module (FAM) is designed to integrate the feature information from different point clouds. Finally, the Assignable Weights Transformer Module (ATM) is presented to stimulate the model’s global modeling ability during the registration process, enabling the selection of representative points for optimal point cloud registration. Extensive experiments conducted on ModelNet40 using partially overlapping point clouds illustrate the superior registration performance of LFA-Net compared with other state-of-the-art methods. Moreover, Numerous experiments on synthetic and real-word datasets further indicate that LFA-Net also has significant advantages in registering partial point clouds with noise and unseen categories, which demonstrates its excellent robustness and generalization ability for real-world practical application.

Chinese Large Language Models (LLMs) have recently demonstrated impressive capabilities across various NLP benchmarks and real-world applications. However, the existing benchmarks for comprehensively evaluating these LLMs are still insufficient, particularly in terms of measuring knowledge that LLMs capture. Current datasets collect questions from Chinese examinations across different subjects and educational levels to address this issue. Yet, these benchmarks primarily focus on objective questions such as multiple-choice questions, leading to a lack of diversity in question types. To tackle this problem, we propose LHMKE, a Large-scale, Holistic, and Multi-subject Knowledge Evaluation benchmark in this paper. LHMKE is designed to provide a comprehensive evaluation of the knowledge acquisition capabilities of Chinese LLMs. It encompasses 10,465 questions across 75 tasks covering 30 subjects, ranging from primary school to professional certification exams. Notably, LHMKE includes both objective and subjective questions, offering a more holistic evaluation of the knowledge level of LLMs. We have assessed 11 Chinese LLMs under the zero-shot setting, which aligns with real examinations, and compared their performance across different subjects. We also conduct an in-depth analysis to check whether GPT-4 can automatically score subjective predictions. Our findings suggest that LHMKE is a challenging and advanced testbed for Chinese LLMs.

Cognitive processing signals can be used to improve natural language processing (NLP) tasks. However, it is not clear how these signals correlate with linguistic information. Bridging between human language processing and linguistic features has been widely studied in neurolinguistics, usually via single-variable controlled experiments with highly-controlled stimuli. Such methods not only compromises the authenticity of natural reading, but also are time-consuming and expensive. In this paper, we propose a data-driven method to investigate the relationship between cognitive processing signals and linguistic features. Specifically, we present a unified attentional framework that is composed of embedding, attention, encoding and predicting layers to selectively map cognitive processing signals to linguistic features. We define the mapping procedure as a bridging task and develop 12 bridging tasks for lexical, syntactic and semantic features. The proposed framework only requires cognitive processing signals recorded under natural reading as inputs, and can be used to detect a wide range of linguistic features with a single cognitive dataset. Observations from experiment results resonate with previous neuroscience findings. In addition to this, our experiments also reveal a number of interesting findings, such as the correlation between contextual eye-tracking features and tense of sentence.

Cohesion devices, e.g., reiteration, coreference, are crucial for building cohesion links across sentences. In this paper, we propose a document-level neural machine translation framework, CoDoNMT, which models cohesion devices from two perspectives: Cohesion Device Masking (CoDM) and Cohesion Attention Focusing (CoAF). In CoDM, we mask cohesion devices in the current sentence and force NMT to predict them with inter-sentential context information. A prediction task is also introduced to be jointly trained with NMT. In CoAF, we attempt to guide the model to pay exclusive attention to relevant cohesion devices in the context when translating cohesion devices in the current sentence. Such a cohesion attention focusing strategy is softly applied to the self-attention layer. Experiments on three benchmark datasets demonstrate that our approach outperforms state-of-the-art document-level neural machine translation baselines. Further linguistic evaluation validates the effectiveness of the proposed model in producing cohesive translations.

Yuqi Ren, Deyi Xiong. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

In this paper, we propose a gated iterative capsule network model for the ADR detection task, named GICN. To alleviate the impact caused by abbreviations and misspelled words, we add character embedding as part of the input. Most ADRs consist of multiple words, e.g., short-term memory dysfunction. Hence, we apply a convolutional neural network (CNN) to obtain the complete phrase information. To effectively extract deep semantic information, we introduce a capsule network with a gated iteration unit that clusters features from underlying to high capsules. The gated iteration mechanism can remember contextual information, which will be introduced when clustering features. Experimental results show that our approach can achieve significant performance improvement for ADR detection from social media text compared with other state-of-the-art works.

The construction of intelligent informatization of the power grid has prompted to accumulate a large amount of text data, and deeply mining the valuable information is significant for the development of the industry. A large number of basic information and information of production process are recorded in the defect text of power equipment. However, repeated expression, unclear logical expression and colloquialism will appear in the process of recording, which makes the operation and maintenance personnel unable to accurately and efficiently manage the logical relationship between the contents of the text. In this paper, we propose a named entity recognition model of BiLSTM-CRF power defect text based on selectable word vector. The model can recognize the category information of professional named entities in the power field from a large number of power defect texts, thereby structuring the massive power defect text data and facilitating the management of text data. In order to further improve the recognition accuracy of the model, we propose an effective optimization scheme. Simulation experiment results show that the proposed model can accurately recognize the information of named entity of defective text in power field. This research can be seen as the foundation for the future defect analysis of power equipment, auxiliary decision support and so on.

BACKGROUND: The adverse reactions that are caused by drugs are potentially life-threatening problems. Comprehensive knowledge of adverse drug reactions (ADRs) can reduce their detrimental impacts on patients. Detecting ADRs through clinical trials takes a large number of experiments and a long period of time. With the growing amount of unstructured textual data, such as biomedical literature and electronic records, detecting ADRs in the available unstructured data has important implications for ADR research. Most of the neural network-based methods typically focus on the simple semantic information of sentence sequences; however, the relationship of the two entities depends on more complex semantic information. METHODS: In this paper, we propose multihop self-attention mechanism (MSAM) model that aims to learn the multi-aspect semantic information for the ADR detection task. first, the contextual information of the sentence is captured by using the bidirectional long short-term memory (Bi-LSTM) model. Then, via applying the multiple steps of an attention mechanism, multiple semantic representations of a sentence are generated. Each attention step obtains a different attention distribution focusing on the different segments of the sentence. Meanwhile, our model locates and enhances various keywords from the multiple representations of a sentence. RESULTS: Our model was evaluated by using two ADR corpora. It is shown that the method has a stable generalization ability. Via extensive experiments, our model achieved F-measure of 0.853, 0.799 and 0.851 for ADR detection for TwiMed-PubMed, TwiMed-Twitter, and ADE, respectively. The experimental results showed that our model significantly outperforms other compared models for ADR detection. CONCLUSIONS: In this paper, we propose a modification of multihop self-attention mechanism (MSAM) model for an ADR detection task. The proposed method significantly improved the learning of the complex semantic information of sentences.