PROTOTYPE IN NEURAL MACHINE TRANSLATION OF PROMOTIONAL TEXTS: ALGORITHMIZATION OF THE SOURCE TEXT

Abstract

The article examines the quality of neural machine translation of promotional texts in the context of digitalization and the active integration of artificial intelligence into multilingual communication. The authors state that source texts should be deliberately optimized for machine translation in order to enhance the quality of translations produced by large language models (LLMs). The empirical material comprises promotional content from the multilingual website of the Singapore-based company Linguise, which develops software solutions for connecting websites to neural machine translation systems, as well as the website of the German hospitality company Traumferienhäuser Schwarzwald – presented by Linguise as a successful case of multilingual logistics. The methodology combines contrastive and translatological analysis of source and target texts, aimed at identifying structural, grammatical, and lexical congruence, as well as typical transformations and translation techniques. The results prove that a high degree of predictability and adequacy in neural machine translation is achieved through the prototypization of the source text, including a modular text structure, a fixed sequence of functional blocks, syntactic simplicity, and terminological consistency. It affirms that texts oriented towards a universal international market and devoid of culturally marked and expressive vocabulary exhibit the greatest stability in automatic translation. Based on the analysis, an algorithm for creating an "ideal" promotional source text is proposed, contributing to improved quality of multilingual neural machine translation and a reduced need for post-editing.