Pacific creole machine translation improves with balanced domain data

The University of Melbourne WMT 2026 CreoleMT Submission: A Domain-Balanced Approach to Low-Resource Pacific Creole Machine Translation

Computation and Language

Summary

Translating Pacific creole languages like Tok Pisin, Bislama, and Solomon Pijin into other languages is hard because there isn’t much data available. The authors trained translation models first on lots of unevenly sourced data, then improved them by fine-tuning with a balanced mix from many types of sources. They used clever data techniques like fixing spelling, generating extra translation pairs, and learning from other models. Their new models translated better than existing open ones on speech and text tests. They plan to create human-written test sets and make smaller, efficient versions retaining their broad coverage.

What this means in practice

  • For language technology developers: Build better translation tools covering diverse content types for low-resource Pacific creole languages using domain-balanced fine-tuned models.
  • For speech recognition engineers: Improve translation of spoken Pacific creole transcripts by integrating models trained on novel spoken language test data.

Authors

Raphaël Merx, Nick Thieberger, Ekaterina Vylomova

Abstract

For our submission to the WMT26 Creole Language Translation Shared Task, we focus on machine translation (MT) models for Pacific creoles: Tok Pisin, Bislama, and Solomon Pijin, with particular attention to broad domain performance. After pre-training on a large collection of domain-imbalanced data, we continue fine-tuning on a diverse mix of domain-balanced data. We rely on a number of data collection and preparation techniques, including LLM-assisted respelling and alignment, back-translation, and distillation from Gemini for domains originally not present in training data. Evaluated on Bouquet and a novel test set made of spoken language transcripts, our models beat open model baselines by 3+ chrF++ points in all directions with human-original references. Looking ahead, we plan to develop human-translated test sets for Solomon Pijin and Bislama, and to distil our best models into much smaller ones that retain broad domain coverage.
View PDFOpen arXiv