Cross-Cultural Sarcasm Detection Using Transformer Models: A Study on Linguistic and Cultural Adaptation
J. Vijaya, Arunkumar Gopu, Heena Painkra, Paridhi Chauhan, Mamta Nag
Abstract
This paper studies transformer models, especially Bidirectional Encoder Representations from Transformers (BERT), to identify sarcasm in text. Detecting sarcasm is difficult because it depends heavily on understanding the specific context, which makes it an implicit challenge. Sarcasm often relies on subtle cues, such as tone, word choice, and contextual references, making it difficult for traditional machine learning models to identify accurately. The primary goal is to enhance the model’s ability to effectively distinguish sarcastic expressions by leveraging BERT’s deep contextual understanding. This capability helps the model grasp the complex connections between words and their context, both from left to right and right to left. This approach is particularly valuable in sarcasm detection, where the meaning often contradicts the literal text. Additionally, the research explores BERT’s cross-cultural adaptability by evaluating its performance on datasets in various languages and cultural contexts. This is achieved through transfer learning, where a model trained on one dataset (e.g., English) is adapted to a new dataset in a different language or cultural context without requiring extensive retraining. The goal is to assess how well BERT can generalize across different social and cultural norms, which may affect how sarcasm is expressed and understood. The findings demonstrate that BERT holds significant potential for cross-cultural sarcasm detection, showing that transformer-based models can be effective in diverse settings.