The increasing prevalence of mental health discourse on social media has created a need for automated tools to assess psychological wellbeing. In this study, we propose a structured framework for evidence extraction, well-being scoring, and summary generation, developed as part of the CLPsych 2025 shared task. Our approach integrates feature-based classification with context-aware language modeling to identify self-state indicators, predict well-being scores, and generate clinically relevant summaries. Our system achieved a recall of 0.56 for evidence extraction, an MSE of 3.89 in well-being scoring, and high consistency scores (0.612 post-level, 0.801 timeline-level) in summary generation, ensuring strong alignment with extracted evidence. With an overall good rank, our framework demonstrates robustness in social media-based mental health monitoring. By providing interpretable assessments of psychological states, our work contributes to early detection and intervention strategies, assisting researchers and mental health professionals in understanding online well-being trends and enhancing digital mental health support systems.

The widespread spread of fake news on social media poses significant challenges, particularly for low-resource languages like Malayalam. The accessibility of social platforms accelerates misinformation, leading to societal polarization and poor decision-making. Detecting fake news in Malayalam is complex due to its linguistic diversity, code-mixing, and dialectal variations, compounded by the lack of large labeled datasets and tailored models. To address these, we developed a fine-tuned transformer-based model for binary and multiclass fake news detection. The binary classifier achieved a macro F1 score of 0.814, while the multiclass model, using multimodal embeddings, achieved a score of 0.1978. Our system ranked 14th and 11th in the shared task competition, highlighting the need for specialized techniques in underrepresented languages. Our full experimental codebase is publicly available at: ciol-researchlab/NAACL25-Akatsuki-Fake-News-Detection.

Language is a cornerstone of cultural identity, yet globalization and the dominance of major languages have placed nearly 3,000 languages at risk of extinction. Existing AI-driven translation models prioritize efficiency but often fail to capture cultural nuances, idiomatic expressions, and historical significance, leading to translations that marginalize linguistic diversity. To address these challenges, we propose a multi-agent AI framework designed for culturally adaptive translation in underserved language communities. Our approach leverages specialized agents for translation, interpretation, content synthesis, and bias evaluation, ensuring that linguistic accuracy and cultural relevance are preserved. Using CrewAI and LangChain, our system enhances contextual fidelity while mitigating biases through external validation. Comparative analysis shows that our framework outperforms GPT-4o, producing contextually rich and culturally embedded translations—a critical advancement for Indigenous, regional, and low-resource languages. This research underscores the potential of multi-agent AI in fostering equitable, sustainable, and culturally sensitive NLP technologies, aligning with the AI Governance, Cultural NLP, and Sustainable NLP pillars of Language Models for Underserved Communities. Our full experimental codebase is publicly avail able at: github.com/ciol-researchlab/Context-Aware_Translation_MAS.

Multilingual emotion detection is a critical challenge in natural language processing, enabling applications in sentiment analysis, mental health monitoring, and user engagement. However, existing models struggle with overlapping emotions, intensity quantification, and cross-lingual adaptation, particularly in low-resource languages. This study addresses these challenges as part of SemEval-2025 Task 11 by leveraging language-specific transformer models for multi-label classification (Track A), intensity prediction (Track B), and cross-lingual generalization (Track C). Our models achieved strong performance in Russian (Track A: 0.848 F1, Track B: 0.8594 F1) due to emotion-rich pretraining, while Chinese (0.483 F1) and Spanish (0.6848 F1) struggled with intensity estimation. Track C faced significant cross-lingual adaptation issues, with Russian (0.3102 F1), Chinese (0.2992 F1), and Indian (0.2613 F1) highlighting challenges in low-resource settings. Despite these limitations, our findings provide valuable insights into multilingual emotion detection. Future work should enhance cross-lingual representations, address data scarcity, and integrate multimodal information for improved generalization and real-world applicability.