Automatic multiple-choice question generation from Thai text

Abstract

This paper presents a method for generating fill-in-the-blank questions with multiple choices from Thai text for testing reading comprehension. The proposed method starts from segmenting input text into clauses by tagging part-of-speech of all words and identifying sentence-breaking spaces. All question phrases are then generated by selecting every tagged-as-noun word as a possible answer. Then, distractors of a question are retrieved by considering all words having the same category with the answer to be distractors. Finally, all generated question phrases and distractors are scored by linear regression models and then ranked to get the most acceptable question phrases and distractors. Custom dictionary is added as an option of the proposed method. The experiment results showed that 81.32% of question phrases generated when a custom dictionary was utilized was rated as acceptable. However, only 49.32% of questions with acceptable question phrases have at least one acceptable distractor. The results also indicated that the ranking process and a custom dictionary can improve acceptability rate of generated questions and distractors.