This paper presents a knowledge-guided humor generation system, which involves generating humorous text from either a pair of words or a news headline. The proposed approach integrates structured semantic reasoning derived from a knowledge graph (KG) with controlled generation using large language models (LLMs). The system constructs an intermediate KG hint consisting of related concepts retrieved in the target language, which is appended to the prompt to guide the generation process in a structured manner. A single candidate joke is generated per input using controlled top-p decoding. Experimental results show that incorporating KG reasoning improves relevance and constraint satisfaction, while LLM-based generation ensures fluency and creativity. Overall, the proposed method offers a structured and interpretable framework for humor generation across multiple languages.

In this paper we present the system for Track A in the SemEval-2024 Task 1: Semantic Textual Relatedness for African and Asian Languages (STR). The proposed system integrates a Siamese Network architecture with pre-trained language models, including BERT, RoBERTa, and the Universal Sentence Encoder (USE). Through rigorous experimentation and analysis, we evaluate the performance of these models across multiple languages. Our findings reveal that the Universal Sentence Encoder excels in capturing semantic similarities, outperforming BERT and RoBERTa in most scenarios. Particularly notable is the USE’s exceptional performance in English and Marathi. These results emphasize the importance of selecting appropriate pre-trained models based on linguistic considerations and task requirements.