CUNI Submission in WMT22 General Task

TL;DR

This paper presents the CUNI-Bergamot submission for WMT22, exploring block backtranslation and MBR decoding techniques to improve English-Czech translation quality, demonstrating their effectiveness individually and combined.

Contribution

It introduces the use of block backtranslation and MBR decoding in translation, showing their combined benefits for the first time.

Findings

Both techniques improve translation quality.

Combining techniques yields better results.

Performance measured with COMET score and entity accuracy.

Abstract

We present the CUNI-Bergamot submission for the WMT22 General translation task. We compete in English$\rightarrow$Czech direction. Our submission further explores block backtranslation techniques. Compared to the previous work, we measure performance in terms of COMET score and named entities translation accuracy. We evaluate performance of MBR decoding compared to traditional mixed backtranslation training and we show a possible synergy when using both of the techniques simultaneously. The results show that both approaches are effective means of improving translation quality and they yield even better results when combined.