Sarcasm detection in text presents a significant challenge for natural language processing (NLP) due to its reliance on nuanced contextual indications rather than straightforward lexical meaning. Existing sentiment analysis approaches often misclassify sarcastic expressions, hindering accurate interpretation of textual communication due to complexities of sarcasm, including ambiguity, sentiment reversal, and the absence of explicit markers. Keyword-based methods are fooled by subtle shifts in tone and machine learning models fail to capture the deeper contextual understanding required. This paper addresses existing challenges by presenting a novel text-based sarcasm detection model leveraging the BERT-base multilingual model, fine-tuned on the Helinivan Sarcasm Headline Multilingual Dataset, enabling the model to learn cross-lingual sarcastic patterns. The proposed model effectively captures the subtle nuances of sarcastic statements by employing contextual embedding and transformer-based sequence classification. Further, the proposed model demonstrated a significant increase in accuracy from 72% to 93% over existing deep learning models such as LSTM, CNN etc. Our contribution lies in enhancing cross-lingual sarcasm detection, improving the robustness of sentiment analysis models, and enabling more accurate interpretation of textual data in applications like social media monitoring, chatbot development, and cross-cultural communication. This research advances the field of NLP by providing a more effective and nuanced approach to understanding sarcasm in diverse linguistic contexts.