Anastasia V. Bubenchikova

The article presents a comprehensive analysis of the limitations of automatic speech translation systems in business communication and examines the risks associated with their use. The theoretical framework is based on translation studies, computational linguistics, cognitive approaches to translation, and psychology. The empirical basis includes comparative studies of machine translation quality, particularly error analysis of DeepL and Google Translate, as well as data from peer-reviewed studies across different language pairs. The study employs a combined methodology including comparative analysis, error classification, and quantitative assessment. It is established that modern neural network systems exhibit consistent lexico-semantic, syntactic, pragmatic, and culture-specific errors, which in business communication translate into legal, reputational, and financial risks. The key difference between professional interpreting and automatic speech translation lies in the human capacity for cognitively responsible interpretation — the integration of semantic, emotional, cultural, and strategic parameters of communication with awareness of the consequences of decisions made. To minimize risks, it is necessary to introduce post-editing standards, develop users’ machine translation literacy, and form hybrid models.