LLM-Based Automatic Generation of Multiple-Choice Questions With Meaningful Distractors

Creating effective Multiple-Choice Questions (MCQs) with high-quality distractors that carefully evaluate students is challenging. Large Language Models (LLMs) can contribute by automatically generating questions. This study presents a novel framework for generating distractors for Portugueselanguage. We evaluate the framework using Sabía-3 (Portuguese-specific) and GPT4o-mini (multilingual). We assessed the grammatical and semantic diversity of the generated distractors and devised a qualitative evaluation through Claude-3 Haiku. Results show that integrating educational principles into prompts enhances the relevance and diversity of distractors, marking progress in automated activity and assessment generation for the Portuguese language.