Skip to content
← All research

Sequence labelling / 02 / Part-of-speech tagging

BalPOS.

A transformer-based Balochi part-of-speech tagger using 16 Universal Dependencies tags.

The challenge

Part-of-speech tags describe how words function in sentences. Balochi needs well-documented tagging tools for language research and downstream NLP.

Approach

  1. 01Prepare labelled examples using Universal Dependencies tag conventions.
  2. 02Fine-tune transformer representations for token classification.
  3. 03Evaluate held-out predictions and report multiple metrics.

Documented outcome

The BalPOS v2 repository reports 87.26% held-out accuracy and 87.22% weighted F1.

Interpretation & limitations

These values describe the reported held-out evaluation dataset, not performance on every dialect, domain or text style.