Analysis of automatic machine translation post-edition techniques

Abstract

Present study concerns automatic post-edition techniques employed in machine translation, voice recognition and other NLP tasks, in the Spanish to Polish language direction. With the aim to find an optimal technical solution for the chosen language pair, which to the best of author’s knowledge does not exist at the time of writing present paper, and significantly lower the costs of translation projects by reducing time and effort required from professional translators, while boosting overall quality, with a state-of-art system, the project focuses on general and medical domains in order to respond to business needs in the translation market and fill in the gap. The undertaken study analyses techniques employed in automatic post-editing architectures, such as Transformers, Bert pretrained language model, and attention mechanism in neural networks, intended at solving linguistic challenges posed by the chosen language systems.