Automatic post-edition (APE) is an architecture designed to correct or modify translations produced by a machine translation (MT) software that translates text from source to target language without human involvement. Both APE and MT have been widely studied from the academic perspective; however, the present study investigates the use of APE in the production context highlighting the business implications it entails. It can significantly reduce both the time and effort required from professional translators, hence cutting the translation cost, while boosting the overall quality. This is especially viable for Romance and Slavic languages with richer morphology than English. To support this claim the research focuses on the state-of-art neural architectures employed in both MT and APE (BERT, fairseq and Transformer) and their applicability to the aforementioned languages. The results obtained (46.89 BLEU) prove that the implementation of APE can be both resource efficient (50% cost reduction) and highly profitable.
Wnuk, Dominika Cecylia and Wołk, Krzysztof, "Analysis of Automatic Post-edition Techniques for Machine Translation Cost Reduction" (2021). PACIS 2021 Proceedings. 98.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.