A SBERT-Based Hybrid Approach for Cognate Identification Between Turkish and Uzbek

This study proposes a hybrid and multi-layered artificial intelligence method for cognate classification to enable a deeper analysis of the historical and semantic relationships between the Turkish and Uzbek languages. The developed model integrates semantic similarity using Sentence-BERT, orthographic similarity using the Levenshtein ratio, and phonetic similarity using the Metaphone algorithm in a holistic approach. Thus, the system provides an objective and reliable comparison based not only on surface similarities but also on the multi-layered analysis of semantic, orthographic, and phonetic dimensions. The main innovation of the study is its ability to address not only definite results but also complex and borderline cases often overlooked by classical methods, through the dynamic evaluation of multiple similarity scores. The system can also identify word pairs that may be considered "uncertain," offering users reasoned and transparent decisions. This holistic approach reduces subjectivity stemming from human interpretation and enables more accurate and traceable analyses of linguistic relationships among Turkic languages. In this way, the model goes beyond automatic cognate classification and objectively reveals linguistic complexity and uncertainty.

Hali tarjima qilinmagan