SLPGFJWUWarda at SemEval-2026 Task 1: A Multimodal Vision-Language Approach for Humor Generation Using Fine-Tuned BLIP
Year: 2026
Type: conference-paper
Abstract: Generating humor requires an AI system to move beyond literal perception and understand shared cultural contexts, irony, and exaggeration.In this paper, we present our submission to SemEval-2026 Task 1 (MWAHAHA), focusing exclusively on Task B1: Image-Based Caption Generation.We propose a multimodal pipeline built upon the Bootstrapping Language-Image Pre-training (BLIP) architecture.By fine-tuning the model on a curated subset of the MemeCap dataset and utilizing beam search decoding, we successfully shift the model outputs from factual descriptions to internet-style comedic captions.Our system efficiently handles animated GIF inputs through first-frame extraction and demonstrates strong contextual humor generation capabilities.
Author: Warda Yousaf
Institution: Fatima Jinnah Women University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166853827