Online polarization, defined as the pronounced division of public opinion into antagonistic groups, poses a significant threat to social cohesion. Automatic detection of polarization across diverse languages and cultures is essential for effective monitoring of online discourse. The challenge extends beyond identifying hate speech to recognizing more nuanced forms, including negative stereotypes, attribution of blame, and dehumanization. This work addresses SemEval-2026 Task 9, which focuses on detecting polarization in multiple languages. Specifically, Subtask 1 involves binary classification of message polarization, while Subtask 2 requires assigning multiple polarization labels in English and Bengali. For Subtask 1, Qwen3-14B is employed with structured few-shot prompting in 4-bit mode, yielding test macro-F1 scores of 0.847 for Bengali (4th place) and 0.808 for English (9th place). For Subtask 2, XLM-RoBERTa-large and RoBERTa-base are fine-tuned using an uneven loss (γ+ = 1, γ− =4) and label-specific thresholds, which increase development macro F1 by up to 24.6 points. The final test macro F1 for English is 0.454 (21st place). Analysis indicates that large language model prompting enhances binary polarization detection, while threshold adjustment is critical for addressing class imbalance in multi-label tasks.

Detecting equivocation is essential, as indirect or evasive responses can shape public perception, influence political narratives, and undermine transparency in democratic discourse. To address the challenge of detecting evasive political responses on digital platforms, participation in the CLARITY SemEval-2026 Task was undertaken, which focuses on (i) clarity-level classification and (ii) fine-grained evasion-type classification in political question-answer contexts. This study introduces a data-centric framework that systematically examines the effects of class distribution and refinement strategies on the performance of Large Language Models (LLMs). A distribution-aware, LLM-augmented dataset was constructed by selectively paraphrasing minority-class instances to enhance class balance, and its performance was benchmarked against full, rebalanced, and undersampled training configurations. To comprehensively assess the proposed method, Qwen3-14B, Phi-4, Gemma-2 9B, and Mistral 7B were evaluated in in-context learning (ICL) settings (zero-shot and few-shot) and with LoRA fine-tuning. Experimental results indicate that fine-tuning Phi-4 with class rebalancing yields strong performance, achieving 74.77% on Subtask-1 and 51.55% on Subtask-2. Consequently, the system ranked 21st in Subtask-1 and 22nd in Subtask-2 on the official evaluation leaderboard.

Language is a rich medium employed to convey emotions subtly and intricately, as abundant as human emotional experiences themselves. Emotion recognition in natural language processing (NLP) is now a core element in facilitating human-computer interaction and interpreting intricate human behavior via text. It has potential applications in every sector i.e., sentiment analysis, mental health surveillance. However, prior research on emotion recognition is primarily from high-resource languages while low-resource languages (LRLs) are not well represented. This disparity has been a limitation to the development of universally applicable emotion detection models. To address this, the SemEval-2025 Shared Task 11 focused on perceived emotions, aiming to identify the emotions conveyed by a text snippet. It includes three tracks: Multi-label Emotion Detection (Track A), Emotion Intensity (Track B), and Cross-lingual Emotion Detection (Track C). This paper explores various models, including machine learning (LR, SVM, RF, NB), deep learning (BiLSTM+CNN, BiLSTM+BiGRU), and transformer-based models (XLM-R, mBERT, ModernBERT). The results showed that XLM-R outperformed other models in Tracks A and B, while BiLSTM+CNN performed better for Track C across most languages.

The rapid growth of digital platforms and social media has significantly contributed to spreading fake news, posing serious societal challenges. While extensive research has been conducted on detecting fake news in high-resource languages (HRLs) such as English, relatively little attention has been given to low-resource languages (LRLs) like Malayalam due to insufficient data and computational tools. To address this challenge, the DravidianLangTech 2025 workshop organized a shared task on fake news detection in Dravidian languages. The task was divided into two sub-tasks, and our team participated in Task 1, which focused on classifying social media texts as original or fake. We explored a range of machine learning (ML) techniques, including Logistic Regression (LR), Multinomial Naïve Bayes (MNB), and Support Vector Machines (SVM), as well as deep learning (DL) models such as CNN, BiLSTM, and a hybrid CNN+BiLSTM. Additionally, this work examined several transformer-based models, including m-BERT, Indic-BERT, XLM-Roberta, and MuRIL-BERT, to exploit the task. Our team achieved 6th place in Task 1, with MuRIL-BERT delivering the best performance, achieving an F1 score of 0.874.

Memes have become one of the main mediums for expressing ideas, humor, and opinions through visual-textual content on social media. The same medium has been used to propagate harmful ideologies, such as misogyny, that undermine gender equality and perpetuate harmful stereotypes. Identifying misogynistic memes is particularly challenging in low-resource languages (LRLs), such as Tamil and Malayalam, due to the scarcity of annotated datasets and sophisticated tools. Therefore, DravidianLangTech@NAACL 2025 launched a Shared Task on Misogyny Meme Detection to identify misogyny memes. For this task, this work exploited an extensive array of models, including machine learning (LR, RF, SVM, and XGBoost), and deep learning (CNN, BiLSTM+CNN, CNN+GRU, and LSTM) are explored to extract textual features, while CNN, BiLSTM + CNN, ResNet50, and DenseNet121 are utilized for visual features.Furthermore, we have explored feature-level and decision-level fusion techniques with several model combinations like MuRIL with ResNet50, MuRIL with BiLSTM+CNN, T5+MuRIL with ResNet50, and mBERT with ResNet50. The evaluation results demonstrated that BERT + ResNet50 performed best, obtaining an F1 score of 0.81716 (Tamil) and were ranked 2nd in the task. The early fusion of MuRIL+ResNet50 showed the highest F1 score of 0.82531 and received a 9th rank in Malayalam.

Text-based hate speech has been prevalent and is usually used to incite hostility and violence. Detecting this content becomes imperative, yet the task is challenging, particularly for low-resource languages in the Devanagari script, which must have the extensive labeled datasets required for effective machine learning. To address this, a shared task has been organized for identifying hate speech targets in Devanagari-script text. The task involves classifying targets such as individuals, organizations, and communities and identifying different languages within the script. We have explored several machine learning methods such as LR, SVM, MNB, and Random Forest, deep learning models using CNN, BiLSTM, GRU, CNN+BiLSTM, and transformer-based models like Indic-BERT, m-BERT, Verta-BERT, XLM-R, and MuRIL. The CNN with BiLSTM yielded the best performance (F1-score of 0.9941), placing the team 13th in the competition for script identification. Furthermore, the fine-tuned MuRIL-BERT model resulted in an F1 score of 0.6832, ranking us 4th for detecting hate speech targets.

Hate speech on social media platforms, particularly in low-resource languages like Bengali, poses a significant challenge due to its nuanced nature and the need to understand its type, severity, and targeted group. To address this, the Bangla Multi-task Hate Speech Identification Shared Task at BLP 2025 adopts a multi-task learning framework that requires systems to classify Bangla YouTube comments across three subtasks simultaneously: type of hate, severity, and targeted group. To tackle these challenges, this work presents BanTriX, a transformer ensemble method that leverages BanglaBERT-I, XLM-R, and BanglaBERT-II. Evaluation results show that the BanTriX, optimized with cross-entropy loss, achieves the highest weighted micro F1-score of 73.78% in Subtask 1C, securing our team 2nd place in the shared task.