Elvira Nurfadhilah, Yuyun, Agung Santosa, Andi Djalal Latief, Dian Isnaeni Nurul Afra, Gusnawaty, Pammuda, Mutahharah Nemin Kaharuddin, Ita Rosvita, Nurfaedah, Hazriani
Part-of-speech (POS) tagging is essential in natural language processing (NLP) that facilitates various downstream applications. However, POS tagging for low-resource languages such as Buginese remains challenging due to the scarcity of annotated data. This paper explores several POS tagging methods, including Unigram, Hidden Markov Model (HMM), Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Gated Recurrent Unit (GRU), integrated with word embedding techniques. We present a comparative analysis of these methods based on their performance on a newly collected and annotated Bugis language dataset. Our results demonstrate that advanced neural models outperform traditional statistical methods, highlighting the potential of deep learning techniques for low-resource language processing, especially RNN, which gets a higher and significant average F1 Score of 97.54%. © 2024 IEEE.
Research Center for Data and Information Sciences, National Agency for Research and Innovation, Jakarta, Indonesia; Department of Regional Language and Literature, Hasanuddin University, Makassar, Indonesia; Department of Indonesian Language and Literature, State University of Makassar, Makassar, Indonesia; Department of Informatics Management, Handayani University, Makassar, Indonesia; Department of Computer System, Handayani University, Makassar, Indonesia
