Building an Elementary Indonesian Textbook Readability Baseline Model

Abstract

Reading resources are crucial for teachers to improve students comprehension on a subject. Readability is a crucial concern for professionals and academics in various sectors and areas of interest, such as education, applied linguistics, text linguistics, library science, business, medical, and technical communication. The readability of texts, readability models, and related techniques have been the subject of numerous studies. In this study, we build a baseline readability model on Indonesian elementary school textbook. We tackled the task as a classification task, which classifies the text into three classes: grade 1, grade 2, and grade 4 that represent three elementary grades. We used traditional predictors or surface-based predictors, syllable patterns, and BERT-based model (IndoBERT) with Random Forest and Support Vector Machine (SVM) as classification methods. The results showed that the IndoBERT neural representation with Support Vector Machine strategy outperforms other scenario with the accuracy of 73.4%.