Dragana Zundanović
- Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Kneza Trpimira 2B, HR-31000 Osijek, Croatia
Dragana Bozic Lenard
Krešimir Romić
- Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Kneza Trpimira 2B, HR-31000 Osijek, Croatia
Hrvoje Leventić
- Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Kneza Trpimira 2B, HR-31000 Osijek, Croatia
Abstract:
Artificial intelligence (AI) has rapidly become indispensable, with applications spanning natural language processing, image generation, and content creation. While large language models (LLMs) primarily process text in English as the global lingua franca, most of the world’s population communicates in other languages, raising questions about the role of low-resource languages in prompt engineering. This paper examines whether the choice of input language—English, Croatian, or a hybrid of both—affects prompt quality and creativity in downstream tasks. Croatian, with its limited digital resources and smaller speaker base, was selected as a case study. A dataset of 100 articles from a Croatian agricultural news portal was used to generate Midjourney prompts via GPT-4o under three language conditions. The prompts and resulting images were analyzed for creativity, semantic consistency, and output quality. Findings reveal subtle but meaningful differences, suggesting that linguistic variation can influence AI-driven creativity. The study underscores the importance of evaluating low-resource languages in multimodal AI systems.
International Scientific Multidisciplinary Conference: AI for a Smarter Tomorrow - AI-SMART , September 25-26, 2025
Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-Non-Commercial 4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and distribution of the work without further permission.
