Search for a command to run...
Large language models have revolutionized various sectors, including education and healthcare, by demonstrating significant advancements in producing human-like text and enhancing translation accuracy. Despite these advancements, the conventional process of cross-cultural instrument translation remains labor-intensive, expensive, and heavily dependent on the expertise of human professionals, including translators and healthcare specialists. The objective of this study is to address the translation challenges in health-related instruments. This study introduces a Human and Large Language Models Collaboration Framework, called H-LLMCF, designed to automate the key steps of instrument translation while maintaining linguistic precision and cultural appropriateness. The proposed framework integrates human expertise and LLM capabilities across two phases: AI-guided translation and human-guided evaluation and refinement. When applied to the translation of two validated instruments in the health sector from English to Arabic, the framework demonstrated promising outcomes. The proposed framework achieved BLEU scores of ~0.54 and ~66 for the two instruments. SBERT-based cosine similarity exceeded 0.85 for 80% of items, with a mean of ~0.92 and ~0.96, indicating strong semantic alignment with human translations. Human evaluation of the 35-item HPV Knowledge Assessment instrument yielded an average I-CVI of 1.00, with 91.42% of items rated as fully acceptable by both reviewers. For the KIDMED instrument, out of 16 items, 15 (93.75%) received a rating of 3 from both reviewers, resulting in an I-CVI of 1.00 for those items. Most discrepancies were linked to translation and fluency adaptation. This highlights the potential of LLMs to enhance cross-cultural translation processes while maintaining translation quality.