A hybrid finite-state machine and small language model architecture for context-aware synonym substitution in Uzbek

This article presents a hybrid architecture developed to solve the problem of context-aware synonym replacement in the Uzbek language. The proposed system combines a finite-state machine and the Qwen model as a small language model. The system is implemented on the Flask API platform and relies on a specially developed semantic dataset containing more than 1100 Uzbek words. The finite-state machine architecture sequentially performs the stages of tokenization, dictionary lookup, context analysis, decision-making, and replacement. Considering the agglutinative nature of the Uzbek language, morphological lemmatization and phonetic reconstruction modules are also integrated.

Hali tarjima qilinmagan