Image Question-Distractors Generation as a Conversational Model

This paper introduces an approach leveraging Visual Language Models (VLM) to automate the generation of educational assessment content for primary school students in Singapore. By integrating Image Captioning and Text Generation tasks, the proposed model generates multiple choice questions (MCQs) from cartoon images aimed at enhancing literacy and comprehension skills. Leveraging advancements in AI, particularly generative AI, the model reduces the workload of educators by automating question-answer generation, thus allowing more efficient use of classroom time. This approach not only improves interactive learning but also addresses the scalability of personalised education through technology. Initial results indicate that the model performs well in generating relevant and contextually appropriate questions. While the model performance is not as good as GPT-40, it can perform inference with significantly fewer resources and limited training examples. This indicates potential performance improvement when more training examples are used.