Syntax-based data augmentation for Hungarian-English machine translation

TL;DR

This paper explores syntax-based data augmentation techniques to improve Transformer-based neural machine translation between Hungarian and English, achieving high BLEU scores and providing publicly available code and models.

Contribution

It introduces a syntax-based augmentation method for neural machine translation and reports state-of-the-art results on Hungarian-English translation tasks.

Findings

Best models achieved BLEU scores of 40.0 (Hungarian-English) and 33.4 (English-Hungarian)

Syntax-based augmentation shows promise for improving translation quality

Code and models are publicly available for further research

Abstract

We train Transformer-based neural machine translation models for Hungarian-English and English-Hungarian using the Hunglish2 corpus. Our best models achieve a BLEU score of 40.0 on HungarianEnglish and 33.4 on English-Hungarian. Furthermore, we present results on an ongoing work about syntax-based augmentation for neural machine translation. Both our code and models are publicly available.