Russian Scientists Train AI to Fix Translation Errors
The neural network translates foreign-language texts and then cleans them up for accuracy and clarity.
Russian researchers have trained an artificial intelligence system to correct errors in translated texts. The development was presented at Volgograd State University, where specialists focused on improving overall content quality. While automated translation tools make it easy to translate text quickly, accuracy often remains an issue.
At the university, researchers conducted a study titled “A Translation Studies Framework for Multilingual Presentation Content in the AI Paradigm.” As part of the project, they identified and systematized the most common translation errors and developed a strategy for correcting them.
The research was based on data from the world’s most widely used online translation services. Some of these systems rely on adaptive approaches, while others attempt to replace phrases with similar idiomatic expressions. The Volgograd team identified recurring spelling and punctuation errors, as well as semantic and stylistic problems, including overly literal translations of set expressions.
Editing Before Translation
The researchers proposed a method of pre-editing texts before translation using a neural network. Sentences are simplified and segmented, complex constructions are rewritten as straightforward narrative, and pronouns and abbreviations are expanded. As a result, the translation becomes significantly more accurate. The approach could be useful not only in linguistics, but also in media, the social sciences, and medical and technical fields.
