Multiple-Choice Question Generation in Low Resource Indonesian Language with T5

Question plays a significant role in the teaching-learning process.However, preparing the questions and assessing each question manually is time-consuming.One type of question, Multiple Choice questions (MCQs) has recently become an active field of research.While studies in this field are not relatively new, they have predominantly focused on the English Language.Currently, there is limited research dedicated to generating MCQs in the Indonesian language.In this study, we propose an end-to-end Indonesian Multiple-Choice Question system that utilises the Textto-Text Transfer Transformer (T5) architecture along with exploring Single-Task & Multi-Task.It contains interconnected components: Question Generation (QG), and Distractor Generation (DG).We utilise the pre-trained T5 model and fine-tuned it to our tasks.Moreover, each of the model performance of these models was evaluated by BLEU, ROUGE-L, and an error analysis.In addition, there are 9 different configurations were tested using Humanlevel performance with total of 22 human assessments were involved in this study.The results show that our Indonesian MCQ generation system represents a promising step toward developing high-quality MCQs.The human evaluation of the best-performing configuration yielded the following scores: 1.38/5 (very easy) for difficulty, 3.99/5 (moderate) for clarity, and 2.07/5 (low-to-moderate) for rigidity (where lower is better).