A Comparative Study of Traditional and Neural Approaches for Gujarati Part-of-Speech Tagging in Scientific and Technical Texts
Contributors
Pooja Bhatt
Pawan Wing
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Part-of-Speech (POS) tagging is a crucial step in the process of Natural Language Processing (NLP) that involves assigning grammatical roles to words, which helps facilitate tasks like parsing, machine translation, and information retrieval at a higher level. While there has been tremendous progress in high resource languages, Gujarati still has issues with small number of annotated corpora, especially in limited domains. A detailed comparative study of four approaches for POS tagging—Hidden Markov Model (HMM), Support Vector Machine (SVM), Conditional Random Fields (CRF) and Bidirectional Long Short-Term Memory (BiLSTM) is presented in this paper, with the use of a manually annotated corpus of Gujarati Science and Technology text, which was prepared according to the BIS tagging standards. Training and evaluation of all models were done following the same experimental setup, thus providing a fair comparison. According to the experimental results, deep learning models are capable of learning the long-range contextual dependency well, and BiLSTM achieves the maximum tagging accuracy of 97.1%, followed by CRF (94.8%), SVM (92.3%) and HMM (88.4%). The results show that neural sequence models work well for domain-specific Gujarati text while CRF is a computationally efficient solution for low-resource applications.