Out-of-Vocabulary Handling in Part-of-Speech Tagging

Part-of-speech (POS) tagging is a key preprocessing step for many NLP tasks. Its broad use in education links this study to Sustainable Development Goals in quality education. Yet models still struggle with out-of-vocabulary (OOV) words. This review maps current solutions. The authors screened 1,357 papers (Jan 2014–Jun 2024) from six databases—Mendeley, IEEE Xplore, ACM DL, SpringerLink, ScienceDirect, and ProQuest—and retained 50 high-quality studies. The review shows that the field has entered a maturity phase, with established approaches—including preprocessing strategies, hand-crafted features, and learned features—being used to address OOV words in POS tagging. Nevertheless, challenges remain in coping with language variation and low-resource data, which indirectly affect model accuracy. This review provides insights into how semantic-web-based strategies can be integrated to overcome these issues and offers guidance not only for POS tagging but also for other NLP tasks involving rare or unseen words.