Fadhli Almu’iini Ahda, Aji Prasetya Wibawa, Didik Dwi Prasetya, Danang Arbian Sulistyo, Andrew Nafalski
This study examines the challenges of processing morphologically intricate languages, focusing on the Minangkabau language, a low-resource language characterized by extensive affixation, particularly in phrases related to traditional medicine. The intricacy of Minangkabau's morphology poses considerable challenges for stemming algorithms and machine translation systems. Current methodologies often struggle to preserve accuracy in languages with complex affix systems, particularly under resource constraints. To address these challenges, we propose a three-stage methodology that integrates Enhanced Confix Stripping (ECS) and Neural Machine Translation (NMT): (1) translating Minangkabau words into Indonesian using NMT, (2) employing ECS to derive root words from the Indonesian translations, and (3) back-translating the root words into Minangkabau through NMT. This method attained accuracies of 81.15% for the first translation, 72.77% for ECS-based stemming, and 64.13% for the final back-translation to Minangkabau, indicating a consistent decrease in accuracy attributable to morphological complexity. These findings underscore the need to enhance ECS algorithms and augment linguistic resources to mitigate issues such as over-and under-stemming. This work underscores the importance of advanced language-processing methodologies in safeguarding cultural and linguistic diversity, thereby laying the foundation for reliable, accessible digital tools for regional language processing. © 2026, Politeknik Negeri Padang. All rights reserved.
Electrical Engineering and Informatics, Universitas Negeri Malang, Malang, Indonesia; Informatics Engineering, Institut Teknologi dan Bisnis Asia Malang, Malang, Indonesia; Electrical Engineering, University of South Australia, Australia
