Revealing the impact of synthetic native samples and multi-tasking strategies in Hindi-English code-mixed humour and sarcasm detection
The Research group of Dr. Jasabanta Patro [Department of Data Science & Engineering] developed “an improved AI framework for detecting humour and sarcasm in code-mixed social media text by leveraging a multi-task learning (MTL) architecture enhanced with native-language sample mixing. This combined approach enabled the model to better capture cultural and linguistic cues, leading to significant performance gains—improving humour detection by up to 10.67% and sarcasm detection by up to 12.35%. The study highlights the effectiveness of integrating native-language signals within an MTL setup for building more accurate and context-aware language technologies in multilingual settings. This work has been recently published in EMNLP Findings 2025.” For more details, kindly visit https://aclanthology.org/2025.findings-emnlp.1308/
