A Hybrid Approach Combining CRF and Rule-Based Logic for Accurate Morphosyntactic Analysis

ABSTRACT This paper presents a hybrid approach to developing a grammar checker for the Telugu language, combining Conditional Random Fields (CRF) for part-of-speech (POS) tagging with handcrafted rule-based validation techniques for grammatical error detection. Telugu, being a morphologically rich and agglutinative Dravidian language, poses unique challenges in natural language processing, including complex verb conjugations, gender and honorific agreements, and clause-level tense consistency. Our system first tokenizes the input text and applies a CRF model to assign POS tags. These tags are then used to drive multiple layers of syntactic and semantic validation rules that check for subject-verb agreement, verb form correctness, tense consistency, missing question particles, redundant verbs, and incorrect or missing postpositions. Although the current implementation demonstrates promising results on a variety of sentence structures, it remains a work-in-progress with limitations in handling all linguistic nuances and context-sensitive constructions. Future improvements aim to enhance robustness through expanded rule coverage and deeper integration with morphological analyzers. This work contributes a foundational step toward automated grammar assistance in low-resource Indian languages like Telugu. Index Terms - Telugu Grammar, Natural Language Processing (NLP), Conditional Random Fields (CRF), Part-of-Speech Tagging, Rule-Based Validation, Morphological Analysis, Syntax Checking, Grammar Correction, Subject-Verb Agreement, Tense Consistency, POS Tagging, Computational Linguistics, Low-Resource Languages, Hybrid NLP Systems.