Usman Naseem

Farhan Ahmad Jafri, Mohammad Aman Siddiqui, Surendrabikram Thapa et al.

The detection of hate speech in political discourse is a critical issue, and this becomes even more challenging in low-resource languages. To address this issue, we introduce a new dataset named IEHate, which contains 11,457 manually annotated Hindi tweets related to the Indian Assembly Election Campaign from November 1, 2021, to March 9, 2022. We performed a detailed analysis of the dataset, focusing on the prevalence of hate speech in political communication and the different forms of hateful language used. Additionally, we benchmark the dataset using a range of machine learning, deep learning, and transformer-based algorithms. Our experiments reveal that the performance of these models can be further improved, highlighting the need for more advanced techniques for hate speech detection in low-resource languages. In particular, the relatively higher score of human evaluation over algorithms emphasizes the importance of utilizing both human and automated approaches for effective hate speech moderation. Our IEHate dataset can serve as a valuable resource for researchers and practitioners working on developing and evaluating hate speech detection techniques in low-resource languages. Overall, our work underscores the importance of addressing the challenges of identifying and mitigating hate speech in political discourse, particularly in the context of low-resource languages. The dataset and resources for this work are made available at https://github.com/Farhan-jafri/Indian-Election.

Wenpeng Lu, Sibo Wei, Xueping Peng et al.

By summarizing longer consumer health questions into shorter and essential ones, medical question-answering systems can more accurately understand consumer intentions and retrieve suitable answers. However, medical question summarization is very challenging due to obvious distinctions in health trouble descriptions from patients and doctors. Although deep learning has been applied to successfully address the medical question summarization (MQS) task, two challenges remain: how to correctly capture question focus to model its semantic intention, and how to obtain reliable datasets to fairly evaluate performance. To address these challenges, this paper proposes a novel medical question summarization framework based on entity-driven contrastive learning (ECL). ECL employs medical entities present in frequently asked questions (FAQs) as focuses and devises an effective mechanism to generate hard negative samples. This approach compels models to focus on essential information and consequently generate more accurate question summaries. Furthermore, we have discovered that some MQS datasets, such as the iCliniq dataset with a 33% duplicate rate, have significant data leakage issues. To ensure an impartial evaluation of the related methods, this paper carefully examines leaked samples to reorganize more reasonable datasets. Extensive experiments demonstrate that our ECL method outperforms the existing methods and achieves new state-of-the-art performance, i.e., 52.85, 43.16, 41.31, 43.52 in terms of ROUGE-1 metric on MeQSum, CHQ-Summ, iCliniq, HealthCareMagic dataset, respectively. The code and datasets are available at https://github.com/yrbobo/MQS-ECL.

Yiran Zhang, Mingyang Lin, Mark Dras et al.

Recent research has increasingly focused on the reasoning capabilities of Large Language Models (LLMs) in multi-turn interactions, as these scenarios more closely mirror real-world problem-solving. However, analyzing the intricate reasoning processes within these interactions presents a significant challenge due to complex contextual dependencies and a lack of specialized visualization tools, leading to a high cognitive load for researchers. To address this gap, we present VISTA, an web-based Visual Interactive System for Textual Analytics in multi-turn reasoning tasks. VISTA allows users to visualize the influence of context on model decisions and interactively modify conversation histories to conduct "what-if" analyses across different models. Furthermore, the platform can automatically parse a session and generate a reasoning dependency tree, offering a transparent view of the model's step-by-step logical path. By providing a unified and interactive framework, VISTA significantly reduces the complexity of analyzing reasoning chains, thereby facilitating a deeper understanding of the capabilities and limitations of current LLMs. The platform is open-source and supports easy integration of custom benchmarks and local models.

Usman Naseem, Byoung Chan Lee, Matloob Khushi et al.

A user-generated text on social media enables health workers to keep track of information, identify possible outbreaks, forecast disease trends, monitor emergency cases, and ascertain disease awareness and response to official health correspondence. This exchange of health information on social media has been regarded as an attempt to enhance public health surveillance (PHS). Despite its potential, the technology is still in its early stages and is not ready for widespread application. Advancements in pretrained language models (PLMs) have facilitated the development of several domain-specific PLMs and a variety of downstream applications. However, there are no PLMs for social media tasks involving PHS. We present and release PHS-BERT, a transformer-based PLM, to identify tasks related to public health surveillance on social media. We compared and benchmarked the performance of PHS-BERT on 25 datasets from different social medial platforms related to 7 different PHS tasks. Compared with existing PLMs that are mainly evaluated on limited tasks, PHS-BERT achieved state-of-the-art performance on all 25 tested datasets, showing that our PLM is robust and generalizable in the common PHS tasks. By making PHS-BERT available, we aim to facilitate the community to reduce the computational cost and introduce new baselines for future works across various PHS-related tasks.

Md. Arid Hasan, Prerona Tarannum, Krishno Dey et al.

Large language models (LLMs) have garnered significant interest in natural language processing (NLP), particularly their remarkable performance in various downstream tasks in resource-rich languages. Recent studies have highlighted the limitations of LLMs in low-resource languages, primarily focusing on binary classification tasks and giving minimal attention to South Asian languages. These limitations are primarily attributed to constraints such as dataset scarcity, computational costs, and research gaps specific to low-resource languages. To address this gap, we present datasets for sentiment and hate speech tasks by translating from English to Bangla, Hindi, and Urdu, facilitating research in low-resource language processing. Further, we comprehensively examine zero-shot learning using multiple LLMs in English and widely spoken South Asian languages. Our findings indicate that GPT-4 consistently outperforms Llama 2 and Gemini, with English consistently demonstrating superior performance across diverse tasks compared to low-resource languages. Furthermore, our analysis reveals that natural language inference (NLI) exhibits the highest performance among the evaluated tasks, with GPT-4 demonstrating superior capabilities.

Muskan Garg, Chandni Saxena, Usman Naseem et al.

Interactions among humans on social media often convey intentions behind their actions, yielding a psychological language resource for Mental Health Analysis (MHA) of online users. The success of Computational Intelligence Techniques (CIT) for inferring mental illness from such social media resources points to NLP as a lens for causal analysis and perception mining. However, we argue that more consequential and explainable research is required for optimal impact on clinical psychology practice and personalized mental healthcare. To bridge this gap, we posit two significant dimensions: (1) Causal analysis to illustrate a cause and effect relationship in the user generated text; (2) Perception mining to infer psychological perspectives of social effects on online users intentions. Within the scope of Natural Language Processing (NLP), we further explore critical areas of inquiry associated with these two dimensions, specifically through recent advancements in discourse analysis. This position paper guides the community to explore solutions in this space and advance the state of practice in developing conversational agents for inferring mental health from social media. We advocate for a more explainable approach toward modeling computational psychology problems through the lens of language as we observe an increased number of research contributions in dataset and problem formulation for causal relation extraction and perception enhancements while inferring mental states.

Qixuan Hu, Shuchang Ye, Xumou Zhang et al.

Cognitive behavioural therapy is widely used to help patients understand and manage psychological distress. It is often delivered through spoken conversation, where therapists attend not only to what patients say, but also to how they say it, because these cues can help therapists decide how to respond and adapt treatment. Progress in building AI systems for CBT remains largely limited to text, partly because most available datasets are text based and shareable spoken CBT data are scarce under ethical and privacy constraints. This creates a blind spot because text based models and evaluations cannot capture the mismatch between the transcript and the patient's voice, even though therapists often rely on this mismatch to understand patient distress. We introduce CBT-Audio, a dataset for evaluating patient distress estimation from spoken CBT sessions with audio language models. CBT-Audio contains 1,802 patient turns from 96 publicly available CBT recordings, with turn-level distress labels validated on an experts-annotated subset. We evaluate 10 open source audio language models under three input conditions, where models receive only patient audio, only the transcript, or both audio and transcript. Our results show that audio can provide useful information beyond text, especially when combined with transcripts. Adding audio to transcript input improves distress estimation over using the transcript alone in 8 of 10 model families, with significant gains in 4, and case studies show the clearest benefit when verbal content and vocal delivery diverge. CBT-Audio makes spoken patient behaviour measurable for AI evaluation in CBT-related tasks and supports future work on audio language models for mental health interaction.

Gautam Siddharth Kashyap, Mark Dras, Usman Naseem

Cultural alignment in Large Language Models (LLMs) is essential for producing contextually aware, respectful, and trustworthy outputs. Without it, models risk generating stereotyped, insensitive, or misleading responses that fail to reflect cultural diversity w.r.t Helpful, Harmless, and Honest (HHH) paradigm. Existing benchmarks represent early steps toward cultural alignment; yet, no benchmarks currently enables systematic evaluation of cultural alignment in line with UNESCO's principles of cultural diversity w.r.t HHH paradigm. Therefore, to address this gap, we built Align-Cultura, two-stage pipeline for cultural alignment. Stage I constructs CULTURAX, the HHH-English dataset grounded in the UNESCO cultural taxonomy, through Query Construction, which reclassifies prompts, expands underrepresented domains (or labels), and prevents data leakage with SimHash. Then, Response Generation pairs prompts with culturally grounded responses via two-stage rejection sampling. The final dataset contains 1,500 samples spanning 30 subdomains of tangible and intangible cultural forms. Stage II benchmarks CULTURAX on general-purpose models, culturally fine-tuned models, and open-weight LLMs (Qwen3-8B and DeepSeek-R1-Distill-Qwen-7B). Empirically, culturally fine-tuned models improve joint HHH by 4%-6%, reduce cultural failures by 18%, achieve 10%-12% efficiency gains, and limit leakage to 0.3%.

Zehui Zhao, Laith Alzubaidi, Jinglan Zhang et al.

Deep learning has significantly advanced automatic medical diagnostics and released the occupation of human resources to reduce clinical pressure, yet the persistent challenge of data scarcity in this area hampers its further improvements and applications. To address this gap, we introduce a novel ensemble framework called `Efficient Transfer and Self-supervised Learning based Ensemble Framework' (ETSEF). ETSEF leverages features from multiple pre-trained deep learning models to efficiently learn powerful representations from a limited number of data samples. To the best of our knowledge, ETSEF is the first strategy that combines two pre-training methodologies (Transfer Learning and Self-supervised Learning) with ensemble learning approaches. Various data enhancement techniques, including data augmentation, feature fusion, feature selection, and decision fusion, have also been deployed to maximise the efficiency and robustness of the ETSEF model. Five independent medical imaging tasks, including endoscopy, breast cancer, monkeypox, brain tumour, and glaucoma detection, were tested to demonstrate ETSEF's effectiveness and robustness. Facing limited sample numbers and challenging medical tasks, ETSEF has proved its effectiveness by improving diagnostics accuracies from 10\% to 13.3\% when compared to strong ensemble baseline models and up to 14.4\% improvements compared with published state-of-the-art methods. Moreover, we emphasise the robustness and trustworthiness of the ETSEF method through various vision-explainable artificial intelligence techniques, including Grad-CAM, SHAP, and t-SNE. Compared to those large-scale deep learning models, ETSEF can be deployed flexibly and maintain superior performance for challenging medical imaging tasks, showing the potential to be applied to more areas that lack training data

Jiahao Yuan, Zhiqing Cui, Hanqing Wang et al.

As web platforms evolve towards greater personalization and emotional complexity, conversational agents must transcend superficial empathy to demonstrate identity-aware emotional reasoning. However, existing systems face two limitations: (1) reliance on situation-centric datasets lacking persistent user identity, which hampers the capture of personalized affective nuances; and (2) dependence on opaque, coarse reward signals that hinder development of verifiable empathetic reasoning. To address these gaps, we introduce KardiaBench, a large-scale user-grounded benchmark comprising 178,080 QA pairs across 22,080 multi-turn conversations anchored to 671 real-world profiles. The dataset is constructed via a model-in-the-loop pipeline with iterative rubric-guided refinement to ensure psychological plausibility and persona consistency. This progressive empathy pipeline that integrates user comprehension, contextual reasoning, and emotion perception into conversations, followed by iterative critique and rubric-based refinement to ensure psychological plausibility, emotional fidelity, and persona consistency. Building on this, we propose Kardia-R1, a framework that trains models for interpretable, stepwise empathetic cognition. Kardia-R1 leverages Rubric-as-Judge Empathetic Reinforcement Learning (Rubric-ERL), a GRPO-based method that uses explainable, human-aligned rubric rewards to tightly couple user understanding, emotional inference, and supportive response generation. Extensive experiments across four LLM backbones demonstrate that Kardia-R1 consistently outperforms othet methods in emotion accuracy, empathy, relevance, persona consistency, and safety. Our dataset and model will be released at https://github.com/JhCircle/Kardia-R1.

Saleh Afzoon, MohammadHossein Ahmadi, Usman Naseem et al.

Personalization and contextual coherence are two essential components in building effective persona-grounded dialogue systems. These aspects play a crucial role in enhancing user engagement and ensuring responses are more relevant and consistent with user identity. However, recent studies indicate that open-source large language models (LLMs) continue to struggle to generate responses that are both contextually grounded and aligned with persona cues, despite exhibiting strong general conversational abilities like fluency and naturalness. We present PersoDPO, a scalable preference optimisation framework that uses supervision signals from automatic evaluations of responses generated by both closed-source and open-source LLMs to fine-tune dialogue models. The framework integrates evaluation metrics targeting coherence and personalization, along with a length-format compliance feature to promote instruction adherence. These signals are combined to automatically construct high-quality preference pairs without manual annotation, enabling a scalable and reproducible training pipeline. Experiments on the FoCus dataset show that an open-source language model fine-tuned with the PersoDPO framework consistently outperforms strong open-source baselines and a standard Direct Preference Optimization (DPO) variant across multiple evaluation dimensions.

Usman Naseem, Robert Geislinger, Juan Ren et al.

We present SemEval-2026 Task 9, a shared task on online polarization detection, covering 22 languages and comprising over 110K annotated instances. Each data instance is multi-labeled with the presence of polarization, polarization type, and polarization manifestation. Participants were asked to predict labels in three sub-tasks: (1) detecting the presence of polarization, (2) identifying the type of polarization, and (3) recognizing the polarization manifestation. The three tasks attracted over 1,000 participants worldwide and more than 10k submission on Codabench. We received final submissions from 67 teams and 73 system description papers. We report the baseline results and analyze the performance of the best-performing systems, highlighting the most common approaches and the most effective methods across different subtasks and languages. The dataset of this task is publicly available.

Qi Zhang, Yayi Yang, Chongyang Shi et al.

The rapid growth of social media has caused tremendous effects on information propagation, raising extreme challenges in detecting rumors. Existing rumor detection methods typically exploit the reposting propagation of a rumor candidate for detection by regarding all reposts to a rumor candidate as a temporal sequence and learning semantics representations of the repost sequence. However, extracting informative support from the topological structure of propagation and the influence of reposting authors for debunking rumors is crucial, which generally has not been well addressed by existing methods. In this paper, we organize a claim post in circulation as an adhoc event tree, extract event elements, and convert it to bipartite adhoc event trees in terms of both posts and authors, i.e., author tree and post tree. Accordingly, we propose a novel rumor detection model with hierarchical representation on the bipartite adhoc event trees called BAET. Specifically, we introduce word embedding and feature encoder for the author and post tree, respectively, and design a root-aware attention module to perform node representation. Then we adopt the tree-like RNN model to capture the structural correlations and propose a tree-aware attention module to learn tree representation for the author tree and post tree, respectively. Extensive experimental results on two public Twitter datasets demonstrate the effectiveness of BAET in exploring and exploiting the rumor propagation structure and the superior detection performance of BAET over state-of-the-art baseline methods.

Utsav Maskey, Mark Dras, Usman Naseem

Aligned language models that are trained to refuse harmful requests also exhibit over-refusal: they decline safe instructions that seemingly resemble harmful instructions. A natural approach is to ablate the global refusal direction, steering the hidden-state vectors away or towards the harmful-refusal examples, but this corrects over-refusal only incidentally while disrupting the broader refusal mechanism. In this work, we analyse the representational geometry of both refusal types to understand why this happens. We show that harmful-refusal directions are task-agnostic and can be captured by a single global vector, whereas over-refusal directions are task-dependent: they reside within the benign task-representation clusters, vary across tasks, and span a higher-dimensional subspace. Linear probing confirms that the two refusal types are representationally distinct from the early transformer layers. These findings provide a mechanistic explanation of why global direction ablation alone cannot address over-refusal, and establish that task-specific geometric interventions are necessary.

Yiran Zhang, Jincheng Hu, Mark Dras et al.

Large language models (LLMs) excel at single-turn reasoning but often lose accuracy and coherence over extended, multi-turn interactions. Recent evaluations such as TurnBench highlight recurring failure modes-reasoning bias, task drift, hallucination, overconfidence, and memory decay. Current approaches typically append full conversational histories, causing unbounded context growth, higher computational costs, and degraded reasoning efficiency. We introduce CogMem, a cognitively inspired, memory-augmented LLM architecture that supports sustained iterative reasoning through structured, persistent memory. CogMem incorporates three layers: a Long-Term Memory (LTM) that consolidates cross-session reasoning strategies; a Direct Access (DA) memory that maintains session-level notes and retrieves relevant long-term memories; and a Focus of Attention (FoA) mechanism that dynamically reconstructs concise, task-relevant context at each turn. Experiments on TurnBench show that this layered design mitigates reasoning failures, controls context growth, and improves consistency across extended reasoning chains, moving toward more reliable, human-like reasoning in LLMs.

Zehui Zhao, Laith Alzubaidi, Jinglan Zhang et al.

Recently, transfer learning and self-supervised learning have gained significant attention within the medical field due to their ability to mitigate the challenges posed by limited data availability, improve model generalisation, and reduce computational expenses. Transfer learning and self-supervised learning hold immense potential for advancing medical research. However, it is crucial to recognise that transfer learning and self-supervised learning architectures exhibit distinct advantages and limitations, manifesting variations in accuracy, training speed, and robustness. This paper compares the performance and robustness of transfer learning and self-supervised learning in the medical field. Specifically, we pre-trained two models using the same source domain datasets with different pre-training methods and evaluated them on small-sized medical datasets to identify the factors influencing their final performance. We tested data with several common issues in medical domains, such as data imbalance, data scarcity, and domain mismatch, through comparison experiments to understand their impact on specific pre-trained models. Finally, we provide recommendations to help users apply transfer learning and self-supervised learning methods in medical areas, and build more convenient and efficient deployment strategies.

Jianlong Zhu, Syed Muhammad Jhon Raza Naqvi, Carolin-Theresa Ziemer et al.

Argumentative dialogues across political divides can reduce polarization, yet opportunities for citizens to engage with opposing views in accessible and structured ways remain limited. AI dialogue partners offer a scalable framework for such open-mindedness exercises, but how the format of human-AI dialogues shapes their benefits remains unclear. In a two-session online experiment, 469 US participants were assigned to argue either for or against their own attitude on a contested political issue with an AI chatbot. Our experimental findings show attitude-congruent dialogues produced greater immediate reduction in both affective and opinion polarization than attitude-incongruent dialogues. By contrast, attitude-incongruent dialogues elicited weaker cognitive state empathy than the non-AI reference task but increased cognitive trait empathy in the two-week period between sessions, suggesting the effects of active generation of attitude-incongruent arguments may emerge over time. These findings highlight dialogue design as a key determinant of effective AI-mediated behavioral interventions.

Pengyu Wang, Shuchang Ye, Usman Naseem et al.

Medical report generation (MRG) aims to automatically derive radiology-style reports from medical images to aid in clinical decision-making. However, existing methods often generate text that mimics the linguistic style of radiologists but fails to guarantee clinical correctness, because they are trained on token-level objectives which focus on word-choice and sentence structure rather than actual medical accuracy. We propose a semantic-driven reinforcement learning (SRL) method for medical report generation, adopted on a large vision-language model (LVLM). SRL adopts Group Relative Policy Optimization (GRPO) to encourage clinical-correctness-guided learning beyond imitation of language style. Specifically, we optimise a report-level reward: a margin-based cosine similarity (MCCS) computed between key radiological findings extracted from generated and reference reports, thereby directly aligning clinical-label agreement and improving semantic correctness. A lightweight reasoning format constraint further guides the model to generate structured "thinking report" outputs. We evaluate Medical Report Generation with Sematic-driven Reinforment Learning (MRG-R1), on two datasets: IU X-Ray and MIMIC-CXR using clinical efficacy (CE) metrics. MRG-R1 achieves state-of-the-art performance with CE-F1 51.88 on IU X-Ray and 40.39 on MIMIC-CXR. We found that the label-semantic reinforcement is better than conventional token-level supervision. These results indicate that optimizing a clinically grounded, report-level reward rather than token overlap,meaningfully improves clinical correctness. This work is a prior to explore semantic-reinforcement in supervising medical correctness in medical Large vision-language model(Med-LVLM) training.

Han Wang, Tan Rui Yang, Usman Naseem et al.

Hate speech is a pressing issue in modern society, with significant effects both online and offline. Recent research in hate speech detection has primarily centered on text-based media, largely overlooking multimodal content such as videos. Existing studies on hateful video datasets have predominantly focused on English content within a Western context and have been limited to binary labels (hateful or non-hateful), lacking detailed contextual information. This study presents MultiHateClip1 , an novel multilingual dataset created through hate lexicons and human annotation. It aims to enhance the detection of hateful videos on platforms such as YouTube and Bilibili, including content in both English and Chinese languages. Comprising 2,000 videos annotated for hatefulness, offensiveness, and normalcy, this dataset provides a cross-cultural perspective on gender-based hate speech. Through a detailed examination of human annotation results, we discuss the differences between Chinese and English hateful videos and underscore the importance of different modalities in hateful and offensive video analysis. Evaluations of state-of-the-art video classification models, such as VLM, GPT-4V and Qwen-VL, on MultiHateClip highlight the existing challenges in accurately distinguishing between hateful and offensive content and the urgent need for models that are both multimodally and culturally nuanced. MultiHateClip represents a foundational advance in enhancing hateful video detection by underscoring the necessity of a multimodal and culturally sensitive approach in combating online hate speech.

Shaina Raza, Deepak John Reji, Dora D. Liu et al.

Recommender systems, information retrieval, and other information access systems present unique challenges for examining and applying concepts of fairness and bias mitigation in unstructured text. This paper introduces Dbias, which is a Python package to ensure fairness in news articles. Dbias is a trained Machine Learning (ML) pipeline that can take a text (e.g., a paragraph or news story) and detects if the text is biased or not. Then, it detects the biased words in the text, masks them, and recommends a set of sentences with new words that are bias-free or at least less biased. We incorporate the elements of data science best practices to ensure that this pipeline is reproducible and usable. We show in experiments that this pipeline can be effective for mitigating biases and outperforms the common neural network architectures in ensuring fairness in the news articles.

Hessa Abdulrahman Alawwad, Areej Alhothali, Usman Naseem et al.

Textbook question answering (TQA) is a challenging task in artificial intelligence due to the complex nature of context needed to answer complex questions. Although previous research has improved the task, there are still some limitations in textual TQA, including weak reasoning and inability to capture contextual information in the lengthy context. We propose a framework (PLRTQA) that incorporates the retrieval augmented generation (RAG) technique to handle the out-of-domain scenario where concepts are spread across different lessons, and utilize transfer learning to handle the long context and enhance reasoning abilities. Our architecture outperforms the baseline, achieving an accuracy improvement of 4. 12% in the validation set and 9. 84% in the test set for textual multiple-choice questions. While this paper focuses on solving challenges in the textual TQA, It provides a foundation for future work in multimodal TQA where the visual components are integrated to address more complex educational scenarios. Code: https://github.com/hessaAlawwad/PLR-TQA

Jiahao Yuan, Zixiang Di, Zhiqing Cui et al.

Empathetic response generation necessitates the integration of emotional and intentional dynamics to foster meaningful interactions. Existing research either neglects the intricate interplay between emotion and intent, leading to suboptimal controllability of empathy, or resorts to large language models (LLMs), which incur significant computational overhead. In this paper, we introduce ReflectDiffu, a lightweight and comprehensive framework for empathetic response generation. This framework incorporates emotion contagion to augment emotional expressiveness and employs an emotion-reasoning mask to pinpoint critical emotional elements. Additionally, it integrates intent mimicry within reinforcement learning for refinement during diffusion. By harnessing an intent twice reflect mechanism of Exploring-Sampling-Correcting, ReflectDiffu adeptly translates emotional decision-making into precise intent actions, thereby addressing empathetic response misalignments stemming from emotional misrecognition. Through reflection, the framework maps emotional states to intents, markedly enhancing both response empathy and flexibility. Comprehensive experiments reveal that ReflectDiffu outperforms existing models regarding relevance, controllability, and informativeness, achieving state-of-the-art results in both automatic and human evaluations.

Mehwish Nasim, Sanjeevan Selvaganapathy, Neel Ganapathi Sabhahit et al.

Many benchmarks show that large language models can answer direct questions about culture. We study a different question: do they also change how they speak when culture is only implied by the situation? We evaluate 60 culturally grounded conversational scenarios across five languages in three conditions: a neutral baseline (Prompt A), an explicit cultural instruction (Prompt B), and implicit situational cueing (Prompt C). We score responses on 12 pragmatic features covering deference to authority, individual-versus-group framing, and uncertainty management. We define Pragmatic Context Sensitivity (PCS) as the fraction of the Prompt A->B shift that reappears under Prompt A->C. Across four deployed LLMs and five languages (English, German, Hindi, Nepali, Urdu), the primary stable-only PCS mean is 0.196 (SD = 0.113), indicating that the models recover only about one-fifth of the pragmatic shift they can produce when instructed explicitly. Transfer is strongest for authority-related cues (0.299) and weakest for individual-versus-group framing (0.120). Uncertainty-related behaviour is mixed: hedging density exhibits negative explicit gaps in all five languages, suggesting that alignment training actively suppresses the target behaviour. Because Hindi and Urdu share core grammar yet index distinct cultural communities, we use them as a natural control; a paired analysis finds no reliable baseline difference (t = 0.96, p = 0.339, dz = 0.06), suggesting that models respond primarily to linguistic structure rather than to the cultural associations a language carries. We argue that multilingual cultural pragmatics is an explicit-versus-implicit deployment problem, not only a factual knowledge problem.

Shuchang Ye, Mingyuan Meng, Hao Wang et al.

Anatomical structure labels for chest radiographs are essential for medical image segmentation and a broad range of downstream diagnostic tasks. However, annotating anatomy directly on 2D chest radiographs is labor-intensive and intrinsically ambiguous, as 3D anatomical structures are projected onto a single 2D plane where boundaries may overlap, be occluded, or appear only partially visible. Consequently, existing anatomy-labeled chest radiograph datasets remain limited in scale, anatomy coverage, and label reliability. To address these limitations, we introduce RadGenome-Anatomy, the largest anatomy-labeled chest radiograph dataset, containing over 10 million segmentation masks across 210 anatomical structures in 25,692 studies. It is constructed by projecting large-scale 3D anatomical masks from CT volumes into 2D radiographic space through canonical radiographic geometry. This shifts annotation from directly tracing uncertain 2D boundaries to defining anatomy in volumetric space, where structures that overlap or become partially invisible in radiographs remain spatially separable. As a result, each 2D mask represents the physically grounded projected footprint of a volumetrically defined structure. The scale and broad anatomical coverage of RadGenome-Anatomy, including structures that are overlapping, partially visible, or difficult to delineate directly, enable research on geometric measurements as explicit evidence for chest radiograph interpretation. We demonstrate this by training XAnatomy to predict structure-specific masks and derive clinically relevant measurements, achieving diagnostic accuracies of 96.4%, 95.6%, and 89.2% for cardiomegaly, kyphosis, and scoliosis, respectively.

Sahil Tripathi, Gautam Siddharth Kashyap, Mehwish Nasim et al.

Meme-based social abuse detection is challenging because harmful intent often relies on implicit cultural symbolism and subtle cross-modal incongruence. Prior approaches, from fusion-based methods to in-context learning with Large Vision-Language Models (LVLMs), have made progress but remain limited by three factors: i) cultural blindness (missing symbolic context), ii) boundary ambiguity (satire vs. abuse confusion), and iii) lack of interpretability (opaque model reasoning). We introduce CROSS-ALIGN+, a three-stage framework that systematically addresses these limitations: (1) Stage I mitigates cultural blindness by enriching multimodal representations with structured knowledge from ConceptNet, Wikidata, and Hatebase; (2) Stage II reduces boundary ambiguity through parameter-efficient LoRA adapters that sharpen decision boundaries; and (3) Stage III enhances interpretability by generating cascaded explanations. Extensive experiments on five benchmarks and eight LVLMs demonstrate that CROSS-ALIGN+ consistently outperforms state-of-the-art methods, achieving up to 17% relative F1 improvement while providing interpretable justifications for each decision.

Usman Naseem, Gautam Siddharth Kashyap, Ebad Shabbir et al.

Misalignment in Large Language Models (LLMs) refers to the failure to simultaneously satisfy safety, value, and cultural dimensions, leading to behaviors that diverge from human expectations in real-world settings where these dimensions must co-occur. Existing benchmarks, such as SAFETUNEBED (safety-centric), VALUEBENCH (value-centric), and WORLDVIEW-BENCH (culture-centric), primarily evaluate these dimensions in isolation and therefore provide limited insight into their interactions and trade-offs. More recent efforts, including MIB and INTERPRETABILITY BENCHMARK-based on mechanistic interpretability, offer valuable perspectives on model failures; however, they remain insufficient for systematically characterizing cross-dimensional trade-offs. To address these gaps, we introduce MisAlign-Profile, a unified benchmark for measuring misalignment trade-offs inspired by mechanistic profiling. First, we construct MISALIGNTRADE, an English misaligned-aligned dataset across 112 normative domains taxonomies, including 14 safety, 56 value, and 42 cultural domains. In addition to domain labels, each prompt is classified with one of three orthogonal semantic types-object, attribute, or relations misalignment-using Gemma-2-9B-it and expanded via Qwen3-30B-A3B-Instruct-2507 with SimHash-based fingerprinting to avoid deduplication. Each prompt is paired with misaligned and aligned responses through two-stage rejection sampling to ensure quality. Second, we benchmark general-purpose, fine-tuned, and open-weight LLMs on MISALIGNTRADE-revealing 12%-34% misalignment trade-offs across dimensions.

Saleh Afzoon, Amin Beheshti, Usman Naseem

Understanding and classifying user personas is critical for delivering effective personalization. While persona information offers valuable insights, its full potential is realized only when contextualized, linking user characteristics with situational context to enable more precise and meaningful service provision. Existing systems often treat persona and context as separate inputs, limiting their ability to generate nuanced, adaptive interactions. To address this gap, we present PersoPilot, an agentic AI-Copilot that integrates persona understanding with contextual analysis to support both end users and analysts. End users interact through a transparent, explainable chat interface, where they can express preferences in natural language, request recommendations, and receive information tailored to their immediate task. On the analyst side, PersoPilot delivers a transparent, reasoning-powered labeling assistant, integrated with an active learning-driven classification process that adapts over time with new labeled data. This feedback loop enables targeted service recommendations and adaptive personalization, bridging the gap between raw persona data and actionable, context-aware insights. As an adaptable framework, PersoPilot is applicable to a broad range of service personalization scenarios.

Pranav Bhandari, Usman Naseem, Mehwish Nasim

Personality steering in large language models (LLMs) commonly relies on injecting trait-specific steering vectors, implicitly assuming that personality traits can be controlled independently. In this work, we examine whether this assumption holds by analysing the geometric relationships between Big Five personality steering directions. We study steering vectors extracted from two model families (LLaMA-3-8B and Mistral-8B) and apply a range of geometric conditioning schemes, from unconstrained directions to soft and hard orthonormalisation. Our results show that personality steering directions exhibit substantial geometric dependence: steering one trait consistently induces changes in others, even when linear overlap is explicitly removed. While hard orthonormalisation enforces geometric independence, it does not eliminate cross-trait behavioural effects and can reduce steering strength. These findings suggest that personality traits in LLMs occupy a slightly coupled subspace, limiting fully independent trait control.

Afrozah Nadeem, Agrima Seth, Mehwish Nasim et al.

Large Language Models (LLMs) increasingly shape global discourse, making fairness and ideological neutrality essential for responsible AI deployment. Despite growing attention to political bias in LLMs, prior work largely focuses on high-resource, Western languages or narrow multilingual settings, leaving cross-lingual consistency and safe post-hoc mitigation underexplored. To address this gap, we present a large-scale multilingual evaluation of political bias spanning 50 countries and 33 languages. We introduce a complementary post-hoc mitigation framework, Cross-Lingual Alignment Steering (CLAS), designed to augment existing steering methods by aligning ideological representations across languages and dynamically regulating intervention strength. This method aligns latent ideological representations induced by political prompts into a shared ideological subspace, ensuring cross lingual consistency, with the adaptive mechanism prevents over correction and preserves coherence. Experiments demonstrate substantial bias reduction along both economic and social axes with minimal degradation in response quality. The proposed framework establishes a scalable and interpretable paradigm for fairness-aware multilingual LLM governance, balancing ideological neutrality with linguistic and cultural diversity.

Abid Ali, Diego Molla-Aliod, Usman Naseem

Multimodal summarization requires models to jointly understand textual and visual inputs to generate concise, semantically coherent summaries. Existing methods often inject shallow visual features into deep language models, leading to representational mismatches and weak cross-modal grounding. We propose a unified framework that jointly performs text summarization and representative image selection. Our system, SPeCTrA-Sum (Sampler Perceiver with Cross-modal Transformer and gated Attention for Summarization), introduces two key innovations. First, a Deep Visual Processor (DVP) aligns the visual encoder with the language model at corresponding depths, enabling hierarchical, layer-wise fusion that preserves semantic consistency. Second, a lightweight Visual Relevance Predictor (VRP) selects salient and diverse images by distilling soft labels from a Determinantal Point Processes (DPP) teacher. SPeCTrA-Sum is trained using a multi-objective loss that combines autoregressive summarization, cross-modal alignment, and DPP-based distillation. Experiments show that our system produces more accurate, visually grounded summaries and selects more representative images, demonstrating the benefits of depth-aware fusion and principled image selection for multimodal summarization.

Markus Essl, Marta Moscati, Mubashir Noman et al.

Multimodal sensor fusion has demonstrated remarkable performance improvements over unimodal approaches in 3D object detection for autonomous vehicles. Typically, existing methods transform multimodal data from independent sensors, such as camera and LiDAR, into a unified bird's-eye view (BEV) representation for fusion. Although effective in ideal conditions, this strategy suffers from substantial performance deterioration when camera or LiDAR data are missing, corrupted, or noisy. To address this vulnerability, we develop a framework-agnostic fusion module for camera and LiDAR data that allows for handling cases when one of the two modalities is missing or corrupted. To demonstrate the effectiveness of our module, we instantiate it in BEVFusion [1], a well-established framework to combine camera and LiDAR data for 3D object detection. By means of quantitative experiments on the MultiCorrupt dataset, we demonstrate that our module achieves favorable performance improvements under scenarios of missing and corrupted modalities, substantially outperforming existing unified representation approaches across a wide range of sensor deterioration scenarios and reaching state-of-the-art performance in scenarios of corrupted modality due to extreme weather conditions and sensor failure.

Abid Ali, Diego Molla-Aliod, Usman Naseem

Multimodal Large Language Models (MLLMs) have facilitated Multimodal Summarization with Multimodal Output (MSMO), wherein systems generate concise textual summaries accompanied by salient visuals from multimodal sources. However, current MSMO evaluation remains fragmented: text quality, image-text alignment, and visual diversity are typically assessed in isolation using unimodal metrics, making it difficult to capture whether the modalities jointly support a faithful and useful summary. To address this gap, we introduce MM-Eval, a unified evaluation framework that integrates assessments of textual quality, cross-modal alignment, and visual diversity. MM-Eval comprises three components: (1) text quality, measured using OpenFActScore for factual consistency and G-Eval for coherence, fluency, and relevance; (2) image-text relevance, evaluated via an MLLM-as-a-judge approach; and (3) image-set diversity, quantified using Truncated CLIP Entropy. We calibrate MM-Eval through a learned aggregation model trained on the mLLM-EVAL news benchmark, aligning component contributions with human preferences. Our analysis reveals a text-dominant hierarchy in this setting, where factual consistency acts as a critical determinant of perceived overall quality, while visual relevance and diversity provide complementary signals. MM-Eval improves over heuristic aggregation baselines and provides an interpretable, reference-weak framework for comparative evaluation of multimodal summaries.

Mohammad Anas Azeez, Rafiq Ali, Ebad Shabbir et al.

Large Language Models (LLMs) hold significant promise for transforming digital health by enabling automated medical question answering. However, ensuring these models meet critical industry standards for factual accuracy, usefulness, and safety remains a challenge, especially for open-source solutions. We present a rigorous benchmarking framework using a dataset of over 1,000 health questions. We assess model performance across honesty, helpfulness, and harmlessness. Our results highlight trade-offs between factual reliability and safety among evaluated models -- Mistral-7B, BioMistral-7B-DARE, and AlpaCare-13B. AlpaCare-13B achieves the highest accuracy (91.7%) and harmlessness (0.92), while domain-specific tuning in BioMistral-7B-DARE boosts safety (0.90) despite its smaller scale. Few-shot prompting improves accuracy from 78% to 85%, and all models show reduced helpfulness on complex queries, highlighting ongoing challenges in clinical QA.

Zhiqing Cui, Haotong Xie, Jiahao Yuan et al.

Large language model (LLM) multi-agent systems typically rely on rigid orchestration, committing either to flat per-query routing or to hand-engineered task decomposition, so decomposition depth, worker choice, and inference budget are not jointly optimized under one objective. We introduce Uno-Orchestra, a unified orchestration policy that selectively decomposes a task and dispatches each subtask to an admissible (model, primitive) pair, with both decisions learned together from curated RL trajectories grounded in real worker interactions. Against 22 baselines on a 13-benchmark suite spanning math, code, knowledge, long-context, and agentic tool-use, Uno-Orchestra reaches 77.0% macro pass@1, roughly 16% above the strongest workflow baseline, at roughly an order of magnitude lower per-query cost, advancing the accuracy-efficiency frontier of selective delegation.

Han Wang, Nirmalendu Prakash, Nguyen Khoi Hoang et al.

Topic modeling is a widely used technique for revealing underlying thematic structures within textual data. However, existing models have certain limitations, particularly when dealing with short text datasets that lack co-occurring words. Moreover, these models often neglect sentence-level semantics, focusing primarily on token-level semantics. In this paper, we propose PromptTopic, a novel topic modeling approach that harnesses the advanced language understanding of large language models (LLMs) to address these challenges. It involves extracting topics at the sentence level from individual documents, then aggregating and condensing these topics into a predefined quantity, ultimately providing coherent topics for texts of varying lengths. This approach eliminates the need for manual parameter tuning and improves the quality of extracted topics. We benchmark PromptTopic against the state-of-the-art baselines on three vastly diverse datasets, establishing its proficiency in discovering meaningful topics. Furthermore, qualitative analysis showcases PromptTopic's ability to uncover relevant topics in multiple datasets.

Liang Xiao, Qi Zhang, Chongyang Shi et al.

The proliferation of social media platforms has fueled the rapid dissemination of fake news, posing threats to our real-life society. Existing methods use multimodal data or contextual information to enhance the detection of fake news by analyzing news content and/or its social context. However, these methods often overlook essential textual news content (articles) and heavily rely on sequential modeling and global attention to extract semantic information. These existing methods fail to handle the complex, subtle twists in news articles, such as syntax-semantics mismatches and prior biases, leading to lower performance and potential failure when modalities or social context are missing. To bridge these significant gaps, we propose a novel multi-hop syntax aware fake news detection (MSynFD) method, which incorporates complementary syntax information to deal with subtle twists in fake news. Specifically, we introduce a syntactical dependency graph and design a multi-hop subgraph aggregation mechanism to capture multi-hop syntax. It extends the effect of word perception, leading to effective noise filtering and adjacent relation enhancement. Subsequently, a sequential relative position-aware Transformer is designed to capture the sequential information, together with an elaborate keyword debiasing module to mitigate the prior bias. Extensive experimental results on two public benchmark datasets verify the effectiveness and superior performance of our proposed MSynFD over state-of-the-art detection models.

Krishno Dey, Prerona Tarannum, Md. Arid Hasan et al.

Large Language Models (LLMs) are trained on massive amounts of data, enabling their application across diverse domains and tasks. Despite their remarkable performance, most LLMs are developed and evaluated primarily in English. Recently, a few multi-lingual LLMs have emerged, but their performance in low-resource languages, especially the most spoken languages in South Asia, is less explored. To address this gap, in this study, we evaluate LLMs such as GPT-4, Llama 2, and Gemini to analyze their effectiveness in English compared to other low-resource languages from South Asia (e.g., Bangla, Hindi, and Urdu). Specifically, we utilized zero-shot prompting and five different prompt settings to extensively investigate the effectiveness of the LLMs in cross-lingual translated prompts. The findings of the study suggest that GPT-4 outperformed Llama 2 and Gemini in all five prompt settings and across all languages. Moreover, all three LLMs performed better for English language prompts than other low-resource language prompts. This study extensively investigates LLMs in low-resource language contexts to highlight the improvements required in LLMs and language-specific resources to develop more generally purposed NLP applications.

Pranav Bhandari, Usman Naseem, Amitava Datta et al.

Psychological assessment tools have long helped humans understand behavioural patterns. While Large Language Models (LLMs) can generate content comparable to that of humans, we explore whether they exhibit personality traits. To this end, this work applies psychological tools to LLMs in diverse scenarios to generate personality profiles. Using established trait-based questionnaires such as the Big Five Inventory and by addressing the possibility of training data contamination, we examine the dimensional variability and dominance of LLMs across five core personality dimensions: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. Our findings reveal that LLMs exhibit unique dominant traits, varying characteristics, and distinct personality profiles even within the same family of models.

Miaomiao Ji, Yanqiu Wu, Zhibin Wu et al.

Reward design plays a pivotal role in aligning large language models (LLMs) with human values, serving as the bridge between feedback signals and model optimization. This survey provides a structured organization of reward modeling and addresses three key aspects: mathematical formulation, construction practices, and interaction with optimization paradigms. Building on this, it develops a macro-level taxonomy that characterizes reward mechanisms along complementary dimensions, thereby offering both conceptual clarity and practical guidance for alignment research. The progression of LLM alignment can be understood as a continuous refinement of reward design strategies, with recent developments highlighting paradigm shifts from reinforcement learning (RL)-based to RL-free optimization and from single-task to multi-objective and complex settings.

Kai Fang, Jiangtao Deng, Chengzu Dong et al.

Frequent fluctuations of client nodes in highly dynamic mobile clusters can lead to significant changes in feature space distribution and data drift, posing substantial challenges to the robustness of existing federated learning (FL) strategies. To address these issues, we proposed a mobile cluster federated learning framework (MoCFL). MoCFL enhances feature aggregation by introducing an affinity matrix that quantifies the similarity between local feature extractors from different clients, addressing dynamic data distribution changes caused by frequent client churn and topology changes. Additionally, MoCFL integrates historical and current feature information when training the global classifier, effectively mitigating the catastrophic forgetting problem frequently encountered in mobile scenarios. This synergistic combination ensures that MoCFL maintains high performance and stability in dynamically changing mobile environments. Experimental results on the UNSW-NB15 dataset show that MoCFL excels in dynamic environments, demonstrating superior robustness and accuracy while maintaining reasonable training costs.

Jiahao Yuan, Zixiang Di, Shangzixin Zhao et al.

Large language models (LLMs) face challenges in aligning with diverse cultural values despite their remarkable performance in generation, which stems from inherent monocultural biases and difficulties in capturing nuanced cultural semantics. Existing methods struggle to adapt to unknown culture after fine-tuning. Inspired by cultural geography across five continents, we propose Cultural Palette, a multi-agent framework that redefines cultural alignment as an adaptive "color-blending" process for country-specific adaptation. Our approach harnesses cultural geography across five continents (Africa, America, Asia, Europe, Oceania) through three key steps: First, we synthesize the Pentachromatic Cultural Palette Dataset using GPT-4o, refining continental-level dialogues with Hofstede's cultural dimensions to establish foundational cultural representations. Second, five continent-level alignment agents form specialized cultural communities that generate region-specific draft responses. Third, a Meta Agent employs Cultural MoErges to dynamically blend these cultural "colors" through attention-gated parameter merging, akin to mixing pigments on a palette, resolving conflicts while preserving cultural nuances to produce the final culturally-aligned response. Extensive experiments across various countries demonstrate that Cultural Palette surpasses existing baselines in cultural alignment.

Shangziqi Zhao, Jiahao Yuan, Guisong Yang et al.

Long chain-of-thought (Long-CoT) reasoning improves accuracy in LLMs, yet its verbose, self-reflective style often hinders effective distillation into small language models (SLMs). We revisit Long-CoT compression through the lens of capability alignment and ask: Can pruning improve reasoning? We propose Prune-on-Logic, a structure-aware framework that transforms Long-CoT into logic graphs and selectively prunes low-utility reasoning steps under self-verification constraints. Through systematic analysis across three pruning strategies - targeting entire chains, core reasoning, and verification - we find that verification pruning consistently improves accuracy while reducing token usage, whereas reasoning or indiscriminate pruning degrades performance. Our study reveals that effective pruning aligns supervision with model capacity rather than merely shortening inputs. Gains hold across tasks, model scales, and CoT capability, with larger models benefiting more from pruning due to richer but more redundant reasoning. Our empirical findings highlight pruning as a structural optimization strategy for aligning CoT reasoning with SLM capacity.

Jiahao Yuan, Dehui Du, Hao Zhang et al.

Large language models (LLMs) have shown remarkable performance in reasoning tasks but face limitations in mathematical and complex logical reasoning. Existing methods to improve LLMs' logical capabilities either involve traceable or verifiable logical sequences that generate more reliable responses by constructing logical structures yet increase computational costs, or introduces rigid logic template rules, reducing flexibility. In this paper, we propose Reversal of Thought (RoT), a plug-and-play and cost-effective reasoning framework designed to enhance the logical reasoning abilities of LLMs during the warm-up phase prior to batch inference. RoT utilizes a Preference-Guided Reverse Reasoning warm-up strategy, which integrates logical symbols for pseudocode planning through meta-cognitive mechanisms and pairwise preference self-evaluation to generate task-specific prompts solely through demonstrations, aligning with LLMs' cognitive preferences shaped by RLHF. Through reverse reasoning, we utilize a Cognitive Preference Manager to assess knowledge boundaries and further expand LLMs' reasoning capabilities by aggregating solution logic for known tasks and stylistic templates for unknown tasks. Experiments across various tasks demonstrate that RoT surpasses existing baselines in both reasoning accuracy and efficiency.

Anudeex Shetty, Amin Beheshti, Mark Dras et al.

Alignment techniques have become central to ensuring that Large Language Models (LLMs) generate outputs consistent with human values. However, existing alignment paradigms often model an averaged or monolithic preference, failing to account for the diversity of perspectives across cultures, demographics, and communities. This limitation is particularly critical in health-related scenarios, where plurality is essential due to the influence of culture, religion, personal values, and conflicting opinions. Despite progress in pluralistic alignment, no prior work has focused on health, likely due to the unavailability of publicly available datasets. To address this gap, we introduce VITAL, a new benchmark dataset comprising 13.1K value-laden situations and 5.4K multiple-choice questions focused on health, designed to assess and benchmark pluralistic alignment methodologies. Through extensive evaluation of eight LLMs of varying sizes, we demonstrate that existing pluralistic alignment techniques fall short in effectively accommodating diverse healthcare beliefs, underscoring the need for tailored AI alignment in specific domains. This work highlights the limitations of current approaches and lays the groundwork for developing health-specific alignment solutions.

Abdullah Mohammad, Sushant Kumar Ray, Pushkar Arora et al.

Generative AI-powered by Large Language Models (LLMs)-is increasingly deployed in industry across healthcare decision support, financial analytics, enterprise retrieval, and conversational automation, where reliability, efficiency, and cost control are critical. In such settings, models must satisfy strict constraints on energy, latency, and hardware utilization-not accuracy alone. Yet prevailing evaluation pipelines remain accuracy-centric, creating a Deployment-Evaluation Gap-the absence of operational and economic criteria in model assessment. To address this gap, we present EDGE-EVAL-a industry-oriented benchmarking framework that evaluates LLMs across their full lifecycle on legacy NVIDIA Tesla T4 GPUs. Benchmarking LLaMA and Qwen variants across three industrial tasks, we introduce five deployment metrics-Economic Break-Even (Nbreak), Intelligence-Per-Watt (IPW ), System Density (\r{ho}sys), Cold-Start Tax (Ctax), and Quantization Fidelity (Qret)-capturing profitability, energy efficiency, hardware scaling, serverless feasibility, and compression safety. Our results reveal a clear efficiency frontier-models in the <2B parameter class dominate larger baselines across economic and ecological dimensions. LLaMA-3.2-1B (INT4) achieves ROI break-even in 14 requests (median), delivers 3x higher energy-normalized intelligence than 7B models, and exceeds 6,900 tokens/s/GB under 4-bit quantization. We further uncover an efficiency anomaly-while QLoRA reduces memory footprint, it increases adaptation energy by up to 7x for small models-challenging prevailing assumptions about quantization-aware training in edge deployment.

Hariram Veeramani, Surendrabikram Thapa, Usman Naseem

In the evolving landscape of Environmental, Social, and Corporate Governance (ESG) impact assessment, the ML-ESG-2 shared task proposes identifying ESG impact types. To address this challenge, we present a comprehensive system leveraging ensemble learning techniques, capitalizing on early and late fusion approaches. Our approach employs four distinct models: mBERT, FlauBERT-base, ALBERT-base-v2, and a Multi-Layer Perceptron (MLP) incorporating Latent Semantic Analysis (LSA) and Term Frequency-Inverse Document Frequency (TF-IDF) features. Through extensive experimentation, we find that our early fusion ensemble approach, featuring the integration of LSA, TF-IDF, mBERT, FlauBERT-base, and ALBERT-base-v2, delivers the best performance. Our system offers a comprehensive ESG impact type identification solution, contributing to the responsible and sustainable decision-making processes vital in today's financial and corporate governance landscape.

Pranav Bhandari, Nicolas Fay, Michael Wise et al.

Large Language Models (LLMs) are widely used as conversational agents, exploiting their capabilities in various sectors such as education, law, medicine, and more. However, LLMs are often subjected to context-shifting behaviour, resulting in a lack of consistent and interpretable personality-aligned interactions. Adherence to psychological traits lacks comprehensive analysis, especially in the case of dyadic (pairwise) conversations. We examine this challenge from two viewpoints, initially using two conversation agents to generate a discourse on a certain topic with an assigned personality from the OCEAN framework (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism) as High/Low for each trait. This is followed by using multiple judge agents to infer the original traits assigned to explore prediction consistency, inter-model agreement, and alignment with the assigned personality. Our findings indicate that while LLMs can be guided toward personality-driven dialogue, their ability to maintain personality traits varies significantly depending on the combination of models and discourse settings. These inconsistencies emphasise the challenges in achieving stable and interpretable personality-aligned interactions in LLMs.

Utsav Maskey, Mark Dras, Usman Naseem

This paper presents a systematic evaluation of Large Language Models' (LLMs) behavior on long-tail distributed (encrypted) texts and their safety implications. We introduce a two-dimensional framework for assessing LLM safety: (1) instruction refusal-the ability to reject harmful obfuscated instructions, and (2) generation safety-the suppression of generating harmful responses. Through comprehensive experiments, we demonstrate that models that possess capabilities to decrypt ciphers may be susceptible to mismatched-generalization attacks: their safety mechanisms fail on at least one safety dimension, leading to unsafe responses or over-refusal. Based on these findings, we evaluate a number of pre-LLM and post-LLM safeguards and discuss their strengths and limitations. This work contributes to understanding the safety of LLM in long-tail text scenarios and provides directions for developing robust safety mechanisms.

Afrozah Nadeem, Mark Dras, Usman Naseem

Recent advancements in large language models (LLMs) have enabled their widespread use across diverse real-world applications. However, concerns remain about their tendency to encode and reproduce ideological biases along political and economic dimensions. In this paper, we employ a framework for probing and mitigating such biases in decoder-based LLMs through analysis of internal model representations. Grounded in the Political Compass Test (PCT), this method uses contrastive pairs to extract and compare hidden layer activations from models like Mistral and DeepSeek. We introduce a comprehensive activation extraction pipeline capable of layer-wise analysis across multiple ideological axes, revealing meaningful disparities linked to political framing. Our results show that decoder LLMs systematically encode representational bias across layers, which can be leveraged for effective steering vector-based mitigation. This work provides new insights into how political bias is encoded in LLMs and offers a principled approach to debiasing beyond surface-level output interventions.

Yiran Zhang, Mo Wang, Xiaoyang Li et al.

Despite impressive advances in large language models (LLMs), existing benchmarks often focus on single-turn or single-step tasks, failing to capture the kind of iterative reasoning required in real-world settings. To address this limitation, we introduce TurnBench, a novel benchmark that evaluates multi-turn, multi-step reasoning through an interactive code-breaking task inspired by a "Turing Machine Board Game." In each episode, a model must uncover hidden logical or arithmetic rules by making sequential guesses, receiving structured feedback, and integrating clues across multiple rounds. This dynamic setup requires models to reason over time, adapt based on past information, and maintain consistency across steps-capabilities underexplored in current benchmarks. TurnBench includes two modes: Classic, which tests standard reasoning, and Nightmare, which introduces increased complexity and requires robust inferential chains. To support fine-grained analysis, we provide ground-truth annotations for intermediate reasoning steps. Our evaluation of state-of-the-art LLMs reveals significant gaps: the best model achieves 81.5% accuracy in Classic mode, but performance drops to 17.8% in Nightmare mode. In contrast, human participants achieve 100% in both, underscoring the challenge TurnBench poses to current models. By incorporating feedback loops and hiding task rules, TurnBench reduces contamination risks and provides a rigorous testbed for diagnosing and advancing multi-step, multi-turn reasoning in LLMs.