Terminological Accuracy and Quality Challenges in AI-Assisted Technical Translation: A Critical Review

Abstract

The growing use of neural machine translation and large language models has expanded the possibilities of translating technical and other specialised documents. Yet the grammatical fluency of an automatically generated text does not necessarily establish the accuracy of its domain-specific terminology. This article critically reviews recent research on terminology management and translation quality in AI-assisted technical translation. Drawing on a focused selection of studies published between 2024 and 2026, it distinguishes terminological correctness from broader semantic accuracy, examines the roles of domain context and validated linguistic resources, and considers the limits of model-generated assurances. Rather than proposing or testing a new translation system, the article develops an analytical basis for a compact, terminology-centred quality-assessment framework. The evidence indicates that models can assist with identifying and translating specialist terms, but the reliability of their outputs depends on the domain, translation task and independent verification. The review proposes a five-part assessment framework addressing concept–term equivalence, contextual suitability, documentary consistency, semantic relations and the integrity of technical data and instructions. This framework is a synthesis-based proposal, not an experimentally validated measurement instrument.