Building a Neural Machine Translation System Using Only Synthetic Parallel Data

TL;DR

This paper demonstrates that neural machine translation systems can be effectively built using only synthetic parallel data, introducing a novel pseudo parallel corpus that mixes ground truth and synthetic examples for improved translation quality.

Contribution

It introduces a new pseudo parallel corpus mixing real and synthetic data, enabling NMT training without solely relying on real parallel data.

Findings

Pseudo parallel corpus improves translation quality

Synthetic data alone can train effective NMT systems

Combining real and synthetic data yields the best results

Abstract

Recent works have shown that synthetic parallel data automatically generated by translation models can be effective for various neural machine translation (NMT) issues. In this study, we build NMT systems using only synthetic parallel data. As an efficient alternative to real parallel data, we also present a new type of synthetic parallel corpus. The proposed pseudo parallel data are distinct from previous works in that ground truth and synthetic examples are mixed on both sides of sentence pairs. Experiments on Czech-German and French-German translations demonstrate the efficacy of the proposed pseudo parallel corpus, which shows not only enhanced results for bidirectional translation tasks but also substantial improvement with the aid of a ground truth real parallel corpus.