Published October 18, 2026 | Version 1

IgalaNLP4D v1: Dialect-Aware Igala-English Dataset (1137 sentences)

  • 1. ROR icon Kwararafa University
  • 2. Federal University of applied sciences kaduna

Contributors

Data manager:

  • 1. ROR icon Kwararafa University
  • 2. Federal University of Applied Sciences Kachia, Kaduna State

Description

First Igala-English dataset with dialect tags: 1137 sentences (899 train).

Distribution: Idah 459 (40.4%), Ankpa 297 (26.1%), Ibaji 160 (14.1%), Other 187 (16.4%), Ogugu 32 (2.8%).

Same English yields 20-60 Igala variants. Idah-only model drops 4.7 BLEU on Ibaji (13.2 -> 8.5), mixed model achieves balanced BLEU 11.66, chrF 46.69.

Base: Helsinki-NLP/opus-mt-en-mul (78M). Epoch ablation: 8 epochs converged, 15 overfit.

Model: https://huggingface.co/EtuhE/igala-mt-first
Existing resource: farunawebservices/igala-english-nmt (3,253 sentences, no dialect tags).

Files

Dialect-Stats.csv

Files (94.5 kB)

Name Size
md5:c0c2bf60820fc4538a4f8b38d8fd25f5
1.2 kB Download
md5:30bbef5c269bb6144612cf8bb488d24b
834 Bytes Preview Download
md5:0c104d0a1df00c2083187333658db69b
46.8 kB Preview Download
md5:e60900f9ec6ed71b18df5ca8d76ef3c3
43.0 kB Preview Download
md5:0ffcfc0c513e485b6cac3f5a06b22025
394 Bytes Preview Download
md5:16b3199677d708fe73893aa05a5e1b41
2.3 kB Preview Download

Additional details

Dates

Created
2026-05-13
Dataset created