The rapid spread of misinformation on digital platforms poses a significant threat to the integrity of news. This makes identifying and verifying claims crucial, particularly in multilingual contexts such as Hindi and English. This study evaluates the Description-Aware RoBERTa (DaBERTa) model, which aims to improve claim span identification by integrating domain-specific knowledge through its innovative Description Infuser Network (DescNet). DescNet, comprising the Compositional De-Attention (CoDA) block and the Interactive Gating Mechanism (IGM), enhances the model’s ability to focus on relevant text segments by leveraging tailored claim descriptions, leading to more accurate and context-aware identification of claims. We compared DaBERTa with leading models, including XLM-RoBERTa, Multilingual BERT (mBERT), and Multilingual DeBERTa (mDeBERTa), using the HECSI dataset of 16,000 tweets in English and Hindi. Results show that XLM-RoBERTa achieved the highest F1 score of 0.9171, demonstrating superior performance in multilingual claim span identification. DaBERTa achieved an F1 score of 0.7498, indicating the need for further refinement and additional training. These findings highlight the potential of incorporating description awareness into natural language processing. Future research will focus on expanding training datasets, refining model architectures, and adding contextual features such as sentiment analysis. These efforts aim to develop more reliable tools for multilingual claim span identification, essential for combating misinformation on social media platforms.