Model for automating the translation process using artificial intelligence in the Blackbird integration platform

Abstract

The relevance of this study is driven by the rapid growth of multilingual corporate content and the need to automate translation across heterogeneous systems (e.g., CMS, TMS, code repositories, marketing platforms). Traditional translation automation often relies on isolated use of Translation Management Systems (TMS) or Machine Translation (MT), which creates a gap between content sources, processing workflows, and target publishing systems. Although Large Language Models (LLMs) demonstrate strong translation performance, their direct adoption in production introduces critical issues, including structural integrity violations, loss of tags and markup, terminology instability, and increased manual post-editing. iPaaS-class integration platforms enable orchestration of end-to-end translation processes; however, a formalized automation model for applying LLMs in an iPaaS environment remains insufficiently developed. This work derives a baseline translation automation model and proposes an extended model by adding: an intelligent content routing module, an XLIFF-based structural integrity module, a context-aware LLM processing module, a quality verification module with human-in-the-loop, and a process analytics module. The object of the study is automated translation processes for corporate content in iPaaS-class integration platforms. The subject of the study is methods for constructing an LLM-enabled translation pipeline that ensures document structural integrity, terminology stability, and process scalability within the Blackbird platform environment. Conclusions. The proposed extended model reduces manual post-editing from 38% to 9%, increases document structure preservation accuracy to 98.4%, and decreases terminology errors to 2.6 per 1000 words compared with specialized baseline configurations, including adaptive NMT based on ModernMT Enterprise and an integrated Phrase TMS pipeline. The practical value of this work lies in providing a formalized, scalable model adaptable to different types of corporate content with specific quality and terminology requirements.

Keywords: translation automation, large language models, iPaaS, Blackbird, orchestration, XLIFF, evaluation of results, model, neural networks, machine translation.