Publications
Dataset · Zenodo · August 6, 2026
Uzbek–English MT: Pre-Registered Blind Human Evaluation (corpus, outputs, ratings, scripts)
Ziyoviddin Sirojiddinov, Muftakh Khakimov
Abstract
Evaluation materials accompanying the paper "What Single-Reference BLEU Cannot Show: A Pre-Registered Blind Human Evaluation of Uzbek–English Machine Translation". Contains the frozen 72-sentence corpus (three stylistic layers), 216 system outputs, the blinding key, anonymized human ratings from two independent raters, per-item BLEU/chrF scores, the frozen primary analysis script, and the dated preregistration plan with amendments.
Natural Language Processing TechniquesTranslation Studies and PracticesTopic ModelingBLEU limitationschrFtranslation quality assessment
More by Muftakh Khakimov
- KKT segmentation validity: automatic verb-bounded units vs. UD gold clauses (Uzbek)Zenodo · 2026
- What the Menzerath–Altmann exponent measures: clauses, morphemes and syllables in Uzbek — supplementary packageZenodo · 2026
- Blind pairwise evaluation materials for dictionary-free spelling normalization in a rule-based Uzbek–English MT system (KKT)Zenodo · 2026
- Uzbek MT structure analysis: clause-model hierarchy and measurement-validity diagnostics (code and coded data)Zenodo · 2026