Discussion of methods for improving machine translation<\/p>\n\n\n\n
[Date]<\/strong> 03.00 pm, Friday, December 14th , 2023<\/p>\n\n\n\n Open Lecture: Controllability for English-Ukrainian Machine Translation Based on Specialized Corpora<\/p>\n\n\n\n [Speaker]<\/strong> Dr. Oleksii Turuta [Abstract]<\/strong>
Oleksii Turuta received his PhD in information technology from Kharkiv National University of Radio Electronics (NURE) in 2012 and is associate professor of NURE. He currently focused on multimodal dataanalysis and distillation language models, multimodal MT, paraphrase of texts, controllable answers in Dialog System and Large Language Models (LLMs).<\/p>\n\n\n\n
Human translators usually beat modern machine learning models, as they have much more external context, which can be useful to create a correct translation both from the meaning and style sides. Our purpose \u00a0is to provide a possible solution for the lack of context during machine translation, which would provide an ability to increase the controllability of existing machine translation architectures. We propose a new architecture, which would incorporate this additional embedded context into the translation and compare this new approach to some classic ones like just transfer learning of some new features using an existing, trained model.<\/p>\n\n\n\n