Cognitive metaphors provide a lens for understanding how societies construct and negotiate ideas, including liberty discourse. This study explores conceptual metaphors in liberty discourse by applying a scalable, corpus-driven approach for cognitive analysis. A curated list of thematic keywords related to liberty topics is used to extract relevant sentences from the Corpus of Historical American English (COHA) and the News on the Web (NOW) corpus. MetaPro, a framework grounded in Conceptual Metaphor Theory, processes these sentences to identify metaphorical mappings at scale. Embedding visualizations and frequency counts were applied to both corpora; in COHA, line graphs captured temporal shifts in metaphor usage across time, while in NOW, two-dimensional heatmaps highlighted spatial variation across countries. Selected example phrases illustrate how metaphorical mappings extend across diverse issues and domains. Thus, metaphor distributions and shifts provide a useful empirical lens for identifying changing thematic concerns in liberty discourse, offering a scalable, cognitively grounded method for cultural analysis across time and space. This demonstrates the value of computational methods for large-scale culture research.

Narrative interpretation is an essential aspect of human cognition, enabling individuals to comprehend complex sequences of events, form emotional connections, and engage in nuanced social reasoning. At the heart of this interpretive ability lies emotional understanding, which cognitive scientists often frame through Appraisal Theory, a model that views emotions as the outcome of subjective evaluations of events in relation to goals, values, and beliefs. In this study, we explore whether multimodal large language models (MLLMs) are able to replicate aspects of this human-like narrative and emotional reasoning. Specifically, we examine how well MLLMs interpret visual narratives, with a focus on their ability to identify and appraise emotional content within scenes. We also investigate whether these models can utilize additional narrative descriptions generated by them to enhance their emotional recognition capabilities, as humans often do. To probe these questions, we conducted a series of experiments using two publicly available datasets, EMOTIC and HECO. Contrary to our expectations, our results reveal a consistent and noteworthy pattern: rather than improving the models’ performance, the inclusion of supplementary narrative or contextual information frequently diminishes their ability to accurately recognize emotions. This counterintuitive finding suggests that current MLLMs face significant challenges in integrating multimodal information in a coherent, context-sensitive way. These findings underscore key limitations in the emotional and narrative reasoning capabilities of existing MLLMs and highlight a critical gap between human cognitive processes and current AI approaches.

We present a large-scale study of philosophical cognition through the lens of Conceptual Metaphor Theory. Using a computational metaphor processing system that extracts target concepts, source concepts, and concept mappings from a curated corpus of 50+ canonical texts (300k sentences) spanning ten schools from antiquity to the late twentieth century, we quantify how metaphor organizes philosophical argument. We model temporal dynamics with year-level cosine series, authorial neighborhoods with PCA projections, and school signatures with heatmaps of normalized frequencies. The study demonstrates that the history of philosophy is structured by stable cross-domain schemas that are selectively recombined to address new problems.

Emotion Recognition in Conversation (ERC) focuses on identifying static emotional states, overlooking the cognitive mechanisms that drive emotional transitions. This work introduces a novel emotion prediction task grounded in Appraisal Theory, which conceptualizes emotion as a cognitive evaluation of expectations and their violations. To address this task, we develop a prompt-based reasoning framework that breaks emotional dynamics into three interpretable stages, e.g., expectation inference, violation detection, and emotion-shift prediction, thereby explaining not only which emotion is expressed, but also why it emerges. To examine whether LLMs exhibit human-like affective reasoning, we design six appraisal-informed prompting tasks and evaluate eight representative LLMs across four conversational corpora. A unified two-level evaluation, which measures both emotion classification and transition dynamics, reveals that explicit expectation cues improve accuracy by up to +2.4%, whereas violation-only cues often degrade performance. Our analysis uncovers a robust appraisal pattern across models and datasets: expectation construction is the primary contributor to accurate emotion prediction, while isolated violation cues tend to induce misattribution rather than improve causal reasoning. Beyond label accuracy, transition-level evaluation shows that LLMs capture emotion-shift direction above chance but exhibit a marked stability bias, over-predicting no-change trajectories and under-detecting fine-grained shifts. These findings demonstrate both the promise and the current limits of LLMs in appraisal-driven affective reasoning, and motivate a new cognitively-grounded research direction.

Peer review is essential for the scholarly publishing process. However, its credibility is increasingly brought to questions. Bias is one of the aspects worthy of investigation. Existing research mostly focuses on predefined, explicit bias types, which are insufficient for analyzing the myriad of implicit biases in peer review. Thus, we proposed to study the bias in peer review through the lens of language abstraction, informed by the cognitive theories which suggest that frequency of abstraction in descriptions plays a latent yet important role in bias transmission. Hence, we trained a model to assess the abstraction level of text, and applied it to a review dataset to examine the connection between abstraction and the implicit biases in peer reviews. Results show that there are indeed observable quantitative differences in the abstraction use of reviews recommending to reject versus recommending to accept. Furthermore, reviews for the rejected papers tend to be more abstract than ones for the accepted papers, indicating possible transmission of implicit bias. To the best of our knowledge, our study is the first to study generalized Linguistic Intergroup Bias in the academic text domain.

Large Language Models often generate incorrect factual answers, yet it remains unclear how they internally distinguish facts they know from those they do not. Existing hallucination-detection methods largely rely on behavioral signals or opaque scalar scores, leaving the structure of factual knowledge in hidden state space poorly understood. We address this gap by introducing a coarse-to-fine framework to discover a knowledge boundary: a decision surface in transformer hidden states that separates representations leading to correct factual answers from those leading to errors. Across multiple datasets and model families, we show that factual correctness is predictable from hidden states, and that this signal is concentrated in specific layers and a small number of influential dimensions. Using Qwen3-8B as a case study, we further identify a compact "knob" subspace that retains near-full boundary-prediction performance and supports targeted interventions that bidirectionally modulate the likelihood of the full reference answer under teacher forcing.

Legal general intelligence (GI) refers to artificial intelligence (AI) that encompasses legal understanding, reasoning, and decision-making, simulating the expertise of legal experts across domains. However, existing benchmarks are result-oriented and fail to systematically evaluate the legal intelligence of large language models (LLMs), hindering the development of legal GI. To address this, we propose LexGenius, an expert-level Chinese legal benchmark for evaluating legal GI in LLMs. It follows a Dimension-Task-Ability framework, covering seven dimensions, eleven tasks, and twenty abilities. We use the recent legal cases and exam questions to create multiple-choice questions with a combination of manual and LLM reviews to reduce data leakage risks, ensuring accuracy and reliability through multiple rounds of checks. We evaluate 12 state-of-the-art LLMs using LexGenius and conduct an in-depth analysis. We find significant disparities across legal intelligence abilities for LLMs, with even the best LLMs lagging behind human legal professionals. We believe LexGenius can assess the legal intelligence abilities of LLMs and enhance legal GI development.Our project is available at https://github.com/QwenQKing/LexGenius.

Recently, various excellent and powerful large language models (LLMs) have been utilized to solve a wide range of human problems. However, when faced with complex problems, most users are often unable to provide accurate and effective prompts to interact with LLMs, thus limiting their performance. To address this challenge, we propose Prompt-R1, an end-to-end reinforcement learning framework that utilizes a small-scale LLM (as agent) to collaborate with large-scale LLMs (as environment), replacing users to interact better. This collaboration is presented as a multi-turn interaction, where the small-scale LLM thinks and generates prompts, and the large-scale LLM performs complex reasoning. A double-constrained reward is designed to optimize correctness and quality of generation. Prompt-R1 provides a plug-and-play framework that supports both inference and training with various large-scale LLMs. Experimental results on twelve datasets show that Prompt-R1 significantly outperforms baseline LLMs across various tasks.Our code is available at https://github.com/QwenQKing/Prompt-R1.

Large Language Model (LLM) Agents exhibit inherent reasoning abilities through the collaboration of multiple tools.However, during agent inference, existing methods often suffer from (i) locally myopic generation, due to the absence of lookahead, and (ii) trajectory instability, where minor early errors can escalate into divergent reasoning paths. These issues make it difficult to balance global effectiveness and computational efficiency. To address these two issues, we propose meta-adaptive exploration with LLM agents (MAXS)[https://github.com/exoskeletonzj/MAXS], a meta-adaptive reasoning framework based on LLM Agents that flexibly integrates tool execution and reasoning planning. MAXS employs a lookahead strategy to extend reasoning paths a few steps ahead, estimating the advantage value of tool usage, and combines step consistency variance and inter-step trend slopes to jointly select stable, consistent, and high-value reasoning steps. Additionally, we introduce a trajectory convergence mechanism that controls computational cost by halting further rollouts once path consistency is achieved, enabling a balance between resource efficiency and global effectiveness in multi-tool reasoning. We conduct extensive empirical studies across three base models (MiMo-VL-7B, Qwen2.5-VL-7B, Qwen2.5-VL-32B) and five datasets, demonstrating that MAXS consistently outperforms existing methods in both performance and inference efficiency. Further analysis confirms the effectiveness of our lookahead strategy and tool usage.

Counterfactual reasoning typically involves considering alternatives to actual events. While often applied to understand past events, a distinct form—forward counterfactual reasoning—focuses on anticipating plausible future developments. This type of reasoning is invaluable in dynamic financial markets, where anticipating market developments can powerfully unveil potential risks and opportunities for stakeholders, guiding their decision-making. However, performing this at scale is challenging due to the cognitive demands involved, underscoring the need for automated solutions. Large Language Models (LLMs) offer promise, but remain unexplored for this application. To address this gap, we introduce a novel benchmark, Fin-Force—FINancial FORward Counterfactual Evaluation. By curating financial news headlines and providing structured evaluation, Fin-Force supports LLM based forward counterfactual generation. This paves the way for scalable and automated solutions for exploring and anticipating future market developments, thereby providing structured insights for decision-making. Through experiments on Fin-Force, we evaluate state-of-the-art LLMs and counterfactual generation methods, analyzing their limitations and proposing insights for future research.

Figure-to-Text (F2T) tasks aim to convert structured figure information into natural language text, serving as a bridge between visual perception and language understanding.However, existing evaluation methods remain limited: 1) Reference-based methods can only capture shallow semantic similarities and rely on costly labeled reference text; 2) Reference-free methods depend on multimodal large language models, which suffer from low efficiency and instruction sensitivity; 3) Existing methods provide only sample-level evaluations, lacking interpretability and alignment with expert-level multi-dimensional evaluation criteria.Accordingly, we propose F2TEval, a five-dimensional reference-free evaluation method aligned with expert criteria, covering faithfulness, completeness, conciseness, logicality, and analysis, to support fine-grained evaluation. We design a lightweight mixture-of-experts model that incorporates independent scoring heads and applies the Hilbert-Schmidt Independence Criterion to optimize the disentanglement of scoring representations across dimensions. Furthermore, we construct F2TBenchmark, a human-annotated benchmark dataset covering 21 chart types and 35 application domains, to support research on F2T evaluation. Experimental results demonstrate our model’s superior performance and efficiency, outperforming Gemini-2.0 and Claude-3.5 with only 0.9B parameters.

Data contamination hinders fair LLM evaluation by introducing test data into newer models’ training sets. Existing studies solve this challenge by updating benchmarks with newly collected data. However, they fail to guarantee contamination-free evaluation as the newly collected data may contain pre-existing knowledge, and their benchmark updates rely on intensive human labor. To address these issues, we in this paper propose AntiLeak-Bench, an automated anti-leakage benchmarking framework. Instead of simply using newly collected data, we construct samples with explicitly new knowledge absent from LLMs’ training sets, which thus ensures strictly contamination-free evaluation. We further design a fully automated workflow to build and update our benchmark without human labor. This significantly reduces the cost of benchmark maintenance to accommodate emerging LLMs. Through extensive experiments, we highlight that data contamination likely exists before LLMs’ cutoff time and demonstrate that AntiLeak-Bench effectively overcomes this challenge.

Sustainability reports are key for evaluating companies’ environmental, social and governance (ESG) performance. To analyze these reports, NLP approaches can efficiently extract ESG insights at scale. However, even the most advanced NLP methods lack robustness against ESG content that is greenwashed – i.e. sustainability claims that are misleading, exaggerated, and fabricated. Accordingly, existing NLP approaches often extract insights that reflect misleading or exaggerated sustainability claims rather than objective ESG performance. To tackle this issue, we introduce A3CG - Aspect-Action Analysis with Cross-Category Generalization, as a novel dataset to improve the robustness of ESG analysis amid the prevalence of greenwashing. By explicitly linking sustainability aspects with their associated actions, A3CG facilitates a more fine-grained and transparent evaluation of sustainability claims, ensuring that insights are grounded in verifiable actions rather than vague or misleading rhetoric. Additionally, A3CG emphasizes cross-category generalization. This ensures robust model performance in aspect-action analysis even when companies change their reports to selectively favor certain sustainability areas. Through experiments on A3CG, we analyze state-of-the-art supervised models and LLMs, uncovering their limitations and outlining key directions for future research.

Question answering (QA) tasks serve as a key benchmark for evaluating generation systems. Traditional rule-based metrics, such as accuracy and relaxed-accuracy, struggle with open-ended and unstructured responses. LLM-based evaluation methods offer greater flexibility but suffer from sensitivity to instructions, robustness issues, and high computational costs. To overcome these challenges, we introduce QAEval, a hybrid framework combining rule-based reliability with LLM-based adaptability. QAEval utilizes two high-quality datasets: QAExtract for short-answer extraction and QAScore for scoring model training. By integrating a Mixture of Evaluators model with Dynamic Load Balancing Optimization, QAEval enables accurate, cost-effective QA evaluation. Experimental results show it outperforms models like GPT-4o and Claude-3, achieving 92.3% accuracy with only 0.6B parameters.

Prevailing research concentrates on superficial features or descriptions of images, revealing a significant gap in the systematic exploration of their connotative and aesthetic attributes. Furthermore, the use of cross-modal relation detection modules to eliminate noise from comprehensive image representations leads to the omission of subtle contextual information. In this paper, we present a Visual Connotation and Aesthetic Attributes Understanding Network (Vanessa) for Multimodal Aspect-based Sentiment Analysis. Concretely, Vanessa incorporates a Multi-Aesthetic Attributes Aggregation (MA3) module that models intra- and inter-dependencies among bi-modal representations as well as emotion-laden aesthetic attributes. Moreover, we devise a self-supervised contrastive learning framework to explore the pairwise relevance between images and text via the Gaussian distribution of their CLIP scores. By dynamically clustering and merging multi-modal tokens, Vanessa effectively captures both implicit and explicit sentimental cues. Extensive experiments on widely adopted two benchmarks verify Vanessa’s effectiveness.

Metaphor interpretation is a difficult task in natural language understanding. The development of relevant techniques in this domain is slow, mostly because of the lack of large annotated datasets and effective pre-trained language models (PLMs) for metaphor learning. Thus, we propose a large annotated dataset and a PLM for the metaphor interpretation task. Our foundation model is based on a novel anomalous language modeling (ALM) method, which we benchmark with comparable PLM baselines on the new dataset, finding that it largely improves model performance on metaphor identification and interpretation.

The success of state-of-the-art Natural Language Processing (NLP) systems heavily depends on deep neural networks, which excel in various tasks through strong data fitting and latent feature modeling abilities. However, certain challenges linked to deep neural networks and supervised deep learning deserve considerations, e.g., extensive computing resources, knowledge forgetting, etc. Previous research attempted to tackle these challenges individually through irrelative techniques. However, they do not instigate fundamental shifts in the learning paradigm. In this work, we propose a novel neurosymbolic method for sentiment analysis to tackle these issues. We also propose a novel sentiment-pragmatic knowledge base that places emphasis on human subjectivity within varying domain annotations. We conducted extensive experiments to show that our neurosymbolic framework for sentiment analysis stands out for its lightweight nature, robustness across domains and languages, efficient few-shot training, and rapid convergence.

Recent studies have revealed some issues of Multi-Head Attention (MHA), e.g., redundancy and over-parameterization. Specifically, the heads of MHA were originally designed to attend to information from different representation subspaces, whereas prior studies found that some attention heads likely learn similar features and can be pruned without harming performance. Inspired by the minimum-redundancy feature selection, we assume that focusing on the most representative and distinctive features with minimum resources can mitigate the above issues and lead to more effective and efficient MHAs. In particular, we propose Grouped Head Attention, trained with a self-supervised group constraint that group attention heads, where each group focuses on an essential but distinctive feature subset. We additionally propose a Voting-to-Stay procedure to remove redundant heads, thus achieving a transformer with lighter weights. Extensive experiments are consistent with our hypothesis. Moreover, our method achieves significant performance gains on three well-established tasks while considerably compressing parameters.

Extrapolation reasoning on temporal knowledge graphs (TKGs) aims to forecast future facts based on past counterparts. There are two main challenges: (1) incorporating the complex information, including structural dependencies, temporal dynamics, and hidden logical rules; (2) implementing differentiable logical rule learning and reasoning for explainability. To this end, we propose an explainable extrapolation reasoning framework TEemporal logiCal grapH networkS (TECHS), which mainly contains a temporal graph encoder and a logical decoder. The former employs a graph convolutional network with temporal encoding and heterogeneous attention to embed topological structures and temporal dynamics. The latter integrates propositional reasoning and first-order reasoning by introducing a reasoning graph that iteratively expands to find the answer. A forward message-passing mechanism is also proposed to update node representations, and their propositional and first-order attention scores. Experimental results demonstrate that it outperforms state-of-the-art baselines.

Persona attribute extraction is critical for personalized human-computer interaction. Dialogue is an important medium that communicates and delivers persona information. Although there is a public dataset for triplet-based persona attribute extraction from conversations, its automatically generated labels present many issues, including unspecific relations and inconsistent annotations. We fix such issues by leveraging more reliable text-label matching criteria to generate high-quality data for persona attribute extraction. We also propose a contrastive learning- and generation-based model with a novel hard negative sampling strategy for generalized zero-shot persona attribute extraction. We benchmark our model with state-of-the-art baselines on our dataset and a public dataset, showing outstanding accuracy gains. Our sampling strategy also exceeds others by a large margin in persona attribute extraction.