The Development of the Light Post-editing Module for English-Kazakh Translation
Diana Rakhimova, Vladislav Karyukin, Aidana Karibayeva, Assem Turarbek, Aliya Turganbayeva
Abstract
Applied intelligent systems play an important role in the modern world. One of their tasks is machine translation (MT) from one language into another one. MT allows people to freely communicate despite language barriers. This new technology is a special step in helping to understand what a companion speaks, or writes to you. Automatic post-editing is the task of correcting errors present in texts as a result of machine translation. Since MT cannot always give the desired result, it becomes necessary to edit the translation. The drawbacks of the translation have to be eliminated by post-editing. This need for post-editing is largely determined by the quality of machine translation: low-quality translation leaves a lot of room for post-editing, and high-quality and human translations require minimal text editing. This work describes the development of the light post-editing module for the English-Kazakh language pairs. The neural network model is trained on pairs mt, pe and triplets src, mt, pe using the OpenNMT framework. Then the results of the BLEU metrics mt - pe and mt - ape are compared, and a conclusion about the quality of post-editing is made.