Modeling the Complexity of the Georgian Verbal System through Linked Data and Machine Learning
Elizbarashvili Archil¹, Tsintsadze Magda², Ducassé Mireille³, Khachidze Manana⁴
¹ Tbilisi State University, Tbilisi, Georgia (Email: ) ² Tbilisi State University, Tbilisi, Georgia (Email: ) ³ Université de Rennes, IRISA, Rennes, France (Email: ) ⁴ Tbilisi State University, Tbilisi, Georgia (Email: )
DOI: https://doi.org/10.48545/advance2025-shortpapers-4_2
Abstract
The Georgian verbal system poses significant challenges due to its agglutinative and inflectional nature, lack of infinitives, rich morphological structure, and irregular patterns. To assist Georgian language learners and support lexicographic work, we are developing KARTUVERBS, a comprehensive linked-data platform for Georgian verb forms. This system integrates structured lexical data, semantic web technologies, and machine learning to infer missing lemmas (verbal nouns). Current results show a prediction accuracy of 98% using decision trees, with cross-validation against the Georgian National Corpus. This paper outlines our methodology, the unique linguistic challenges of Georgian, and future directions for integrating crowdsourcing and deep learning into the tool’s development.
Keywords
Linked data, machine learning, Georgian verbs, lexicography, verbal noun, decision tree