Ala Syauqi, Aji Prasetya Wibawa, Anik Nur Handayani
The traditional scriptio continua writing style of Javanese script poses significant challenges for automated transliteration and semantic word segmentation. To address this limitation, this study proposes an interactive human-in-the-loop framework designed for efficient corpus construction. The system integrates a deterministic rule-based transliteration engine with a web-based interface for manual curation. A specialized greedy alignment algorithm featuring a look-ahead mechanism is employed to automatically map curated Latin tokens back to the original Javanese characters. Experimental validation was conducted using a ground truth dataset derived from the 1927 Kejawen periodical. The results demonstrate that the framework achieved a 100% Alignment Success Rate and a negligible processing latency of 0.1 milliseconds per sentence. Additionally, the system successfully extracted 1,992 verified word pairs from 300 sentences. These findings confirm the efficacy of the proposed tool in generating high-quality duallayer parallel corpora, thereby facilitating the development of advanced neural models for low-resource languages. © 2026 IEEE.
Universitas Negeri Malang, Universitas Islam Negeri, Maulana Malik Ibrahim Malang, Faculty of Science and Technology, Department of Electrical Engineering and Informatics, Informatics Engineering, Malang, Indonesia; Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia
