Management of information resources in the scientific and technical sphere: An architectural approach to processing multilingual data arrays

Aim. To develop an architectural approach to managing information resources in the scientific and technical sphere in which multilingualism is treated not as a local machine translation task but as an end-to-end property of data, metadata, terminology, and search and analytical services. Objectives. To define the architectural requirements for integrating multilingual scientific arrays; to form a digital object model with an immutable original, normalized metadata, and a language-independent semantic representation; to propose contours for terminological normalization, entity resolution, and cross-language search; to set criteria for traceability, controlled autonomy, and scenario-based verification. Methods. Systemic and architectural approaches, methods of conceptual modeling of information systems, analysis of the data life cycle, and comparison of contemporary methods of cross-language text representation are used. The methodological foundation consists of Russian and international research on information resources and open science, and works devoted to knowledge graphs, cross-language representations, and models of scientific text. Results. An architecture for managing multilingual scientific and technical resources is developed, consisting of six layers. These include sources and rights; canonization and provenance; linguistic mediation; a language-independent semantic core; search and analytical services; and life cycle management and audit. At its core is a dual representation of the object that preserves the original text alongside the translation. A verification protocol is proposed to test the architecture. The ranking order and an index of cross-language consistency of retrieval results are defined. Conclusions. When managing multilingual arrays, it is necessary to separate the storage of the original, normalization, semantic integration, and presentation of data to the user. Translation remains a derivative projection, with its versions taken into account, and does not serve as the basis for integration. The developed architecture can be used as a methodological foundation for scientific repositories, sectoral scientific and technical information systems, and analytical platforms. It is designed for tasks in which cross-language search must be combined with the preservation of data provenance and the execution of automatic transformations according to controlled rules.