The Role of Lexical and Semantic Features in Shaping Indonesian Academic Text Readability: An NLP-Based Analysis
This study investigates the relationships among lexical features, semantic features, and readability using a corpus of 27,274 sentences extracted from Indonesian undergraduate academic abstracts. A quantitative descriptive-correlational design was employed, combining natural language processing (NLP)-based feature extraction with a regression-calibrated Indonesian adaptation of the Flesch readability formula. Pearson and Spearman correlation analyses were performed to examine associations among lexical density, lexical diversity, lexical bundles, semantic dimensions, and readability scores. The results show that semantic features consistently exhibit stronger relationships with readability than conventional lexical measures. Mean word concreteness exhibited the strongest positive correlation with readability (r = 0.424), whereas the abstractword ratio showed the strongest negative correlation (r = -0.407). In contrast, lexical density, lexical diversity, and lexical bundles displayed only negligible to weak correlations. These findings indicate that semantic clarity has a greater impact on the readability of Indonesian academic writing than lexical sophistication alone. Rather than emphasizing increasingly complex vocabulary, improving readability depends on communicating ideas through concrete and contextually meaningful language. Beyond extending Indonesian readability research, this study provides empirical evidence for the development of NLP-based readability assessment, intelligent academic writing assistance systems, and pedagogical practices that support clearer scholarly communication for national and international audiences.