Morphological Disambiguation in Turkic Languages Based on an Uzbek Homonym Resource
Abstract
Morphological disambiguation is a core preprocessing task for agglutinative languages because one surface form may license several lexical and grammatical analyses. This paper reports two completed contributions for Uzbek: a comparative synthesis of rule-based, statistical, neural, and hybrid approaches, and a descriptive audit of an Uzbek homonym and polysemy resource derived from the Dictionary of Uzbek Homonyms. The resource contains 3,127 sense records for 1,312 normalized word forms; 1,294 forms are multi-sense, with an average of 2.38 senses per form. Binary ambiguity dominates, while 337 forms have three or more senses and therefore require broader contextual evidence. We additionally specify a two-stage architecture in which a finite-state analyzer generates licensed candidates and a context-aware ranker selects among them. The current study does not claim an implemented end-to-end disambiguator, a sentence-level gold corpus, or benchmark Accuracy and F1 results. Its empirical contribution is the resource analysis; the architecture, leakage-controlled evaluation protocol, and reproducibility package define the next experimental stage.