UzAlign v1.0: Uzbek–English Parallel Corpus with Morphological Annotation and Model-Assisted Word Alignment
Year: 2026
Type: dataset
Abstract: UzAlign v1.1 is an updated Uzbek–English parallel corpus developed for research on sentence alignment, morphology-aware natural language processing, and token-level word alignment. The corpus provides Uzbek–English sentence-alignment records, including 1:1, 1:2, and 2:1 alignment configurations. Its linguistic annotation layers cover tokenization, lemmatization, Universal Part-of-Speech (UPOS) tags, and Universal Dependencies morphological features (UD FEATS). Morpheme-level segmentation is provided for Uzbek, while English morphological preprocessing operates at the token and feature levels. Version 1.1 incorporates corrections made following methodological review and includes updated release documentation and supporting audit materials. The word-alignment workflow combines model-assisted alignment proposals with documented human review and correction procedures. The distinction between model-generated proposals, human-reviewed records, and independently validated reference annotations remains important when interpreting the dataset. The accompanying release materials should be consulted for the final dataset statistics, TRAIN/DEV/TEST partitions, annotation procedures, validation evidence, and reproducibility limitations. This release is a revised version of UzAlign v1.0 (DOI: 10.5281/zenodo.22817414).
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7213522322
