Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BBPOS: BERT-based Part-of-Speech Tagging for Uzbek

Latofat Bobojonova, Arofat Akhundjanova, Phil Ostheimer, Sophie Fellenz

TL;DR

This work targets POS tagging for Uzbek, a morphologically rich, low-resource language, by introducing BBPOS—a BERT-based POS tagger trained on two monolingual Uzbek BERTs for Latin and Cyrillic scripts. It also releases the first UPOS-tagged Uzbek benchmark dataset (500 sentences) and demonstrates that monolingual models outperform mBERT and a rule-based tagger, achieving about 91% average accuracy under 5-fold cross-validation. The study shows context sensitivity and the ability to capture affix-driven POS changes, while noting limitations from dataset size and Latin-script tokenization issues. The findings support the value of script-aware, monolingual models for Uzbek NLP and point to morphologically informed tokenization and larger datasets as fruitful directions for future work.

Abstract

This paper advances NLP research for the low-resource Uzbek language by evaluating two previously untested monolingual Uzbek BERT models on the part-of-speech (POS) tagging task and introducing the first publicly available UPOS-tagged benchmark dataset for Uzbek. Our fine-tuned models achieve 91% average accuracy, outperforming the baseline multi-lingual BERT as well as the rule-based tagger. Notably, these models capture intermediate POS changes through affixes and demonstrate context sensitivity, unlike existing rule-based taggers.

BBPOS: BERT-based Part-of-Speech Tagging for Uzbek

TL;DR

This work targets POS tagging for Uzbek, a morphologically rich, low-resource language, by introducing BBPOS—a BERT-based POS tagger trained on two monolingual Uzbek BERTs for Latin and Cyrillic scripts. It also releases the first UPOS-tagged Uzbek benchmark dataset (500 sentences) and demonstrates that monolingual models outperform mBERT and a rule-based tagger, achieving about 91% average accuracy under 5-fold cross-validation. The study shows context sensitivity and the ability to capture affix-driven POS changes, while noting limitations from dataset size and Latin-script tokenization issues. The findings support the value of script-aware, monolingual models for Uzbek NLP and point to morphologically informed tokenization and larger datasets as fruitful directions for future work.

Abstract

This paper advances NLP research for the low-resource Uzbek language by evaluating two previously untested monolingual Uzbek BERT models on the part-of-speech (POS) tagging task and introducing the first publicly available UPOS-tagged benchmark dataset for Uzbek. Our fine-tuned models achieve 91% average accuracy, outperforming the baseline multi-lingual BERT as well as the rule-based tagger. Notably, these models capture intermediate POS changes through affixes and demonstrate context sensitivity, unlike existing rule-based taggers.
Paper Structure (24 sections, 2 figures, 5 tables)

This paper contains 24 sections, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Rule-Based POS Taggers:
  4. Statistical POS Taggers:
  5. Neural POS Taggers:
  6. Dataset & Tagset:
  7. Experiments
  8. Methods
  9. Data
  10. Tagset Selection:
  11. Dataset Development:
  12. Models
  13. Latin BERT:
  14. Cyrillic BERT:
  15. Results
  16. ...and 9 more sections

Figures (2)

  • Figure 1: Analysis of one sentence-word in Uzbek: manual annotation according to UPOS guidelines (top); how BBPOS tags it (middle); comprehensive morphological analysis of the word (bottom).
  • Figure 2: Words per sentence and characters per word/sentence in the dataset.