Parts of Speech (POS) Tagging in Telugu Corpora Using CRF Algorithm

The study of NLP (Natural Language Processing), a branch of computer science and AI (Artificial Intelligence), enables machines to comprehend human language effectively and assist with linguistic tasks. The initial step in every NLP task is POS (Parts of Speech) tagging, which assigns a tag to a word based on its meaning and context. The present paper discusses parts of speech tagging (POS) in Telugu using Conditional Random Fields (CRF), a sequence modelling algorithm that is particularly effective in identifying entities or text patterns, such as POS tags, in highly inflectional and agglutinative languages like Telugu. Telugu is a highly inflectional and agglutinative language widely spoken in the southern part of India (mainly Andhra Pradesh and Telangana). The Language belongs to the Dravidian Family and, it follows the S - O - V structure. Compared to other machine learning algorithms, CRF has been proven more effective in overcoming label - bias problems in a language. In order to understand the language features and to tag the test corpus, an annotated corpus of 62, 996 words and a tag set of 18 tags is used for the study. The present study has achieved an accuracy of 80.17%.