Artificial intelligence as a supportive agent in the crafting of terminology development
The case of Estonian national defence terminology
Bibliographic Data
| ID | 24241330 |
|---|---|
| Authors | Reet Hendrikson (Estonian National Defense College), Eleri Aedmaa (Institute of the Estonian Language) |
| Year | 2026 |
| Publication date | 2026-10-30 |
| Open Access | No |
| Type | ARTICLE |
| Venue | Terminology International Journal of Theoretical and Applied Issues in Specialized Communication (JOURNAL) |
| Journal identifiers | ISSN: 0929-9971 • E-ISSN: 1569-9994 |
| Publisher | John Benjamins Publishing Company (PUBLISHER • NL) |
| DOI | 10.1075/term.00096.hen |
| Language | EN |
| References cited | 22 |
This paper examines the role of artificial intelligence (AI) as a supportive agent in professional terminology work, using the Estonian national defence terminology as a case study. The study addresses the growing need to manage large volumes of multilingual defence-related documentation while simultaneously preserving expert-driven, concept-oriented terminology development practices. We present and evaluate a retrieval-augmented, large language model (LLM)-based prototype designed to extract and structure source-grounded (English) terminological raw material — such as definition candidates, related terms, usage contexts, and exact references — rather than to generate dictionary-ready entries. The prototype was evaluated through an expert-based assessment involving both domain specialists and experienced terminologists. The evaluation focused on the practical usefulness, contextual adequacy, and completeness of the extracted information, rather than on formal definitional correctness alone. The results indicate that the prototype performs reliably in retrieving definition candidates, contextual evidence, and source references, and can meaningfully support the early stages of terminology work. At the same time, persistent limitations were identified, including occasional hallucinations under weak source constraints, inconsistent handling of semantic relations, and challenges at the boundary between general and specialised language. The findings underscore the continued centrality of human expertise in terminology development and support a human-at-the-core model — as distinct from a human-in-the-loop model, where humans primarily correct machine output — in which AI functions as an assistive, context-sensitive tool rather than an autonomous terminological agent. The study contributes contextualised empirical evidence to ongoing debates on AI-assisted terminology management and offers methodological insights into prompt design, expert evaluation, and the integration of AI tools into institutional terminology workflows, particularly in small-language and national defence contexts.
| Highly cited | No |
|---|