Tehisintellekti rakendamine erialakeele arendamisel riigikaitseterminoloogia näitel

Abstract

https://doi.org/10.5128/ERYa21.02

Suurte keelemudelite ja tehisintellekti kiire areng pakub võimalusi nende mitmekülgseks rakendamiseks, mh erialakeele arendamisel. Artikli eesmärk on analüüsida riigikaitsevaldkonna terminitöö näitel välistoelise genereerimise (RAG) sobivust ja tõhusust definitsioonide ja allikaviidete väljatoomisel alusdokumentidest. Esialgsed tulemused näitavad, et RAG-süsteem leiab alusandmetest täpselt definitsioone ja allikaviiteid, mis teeb sellest sobiliku abivahendi terminoloogi töö tõhustamiseks. Seega on alust eeldada, et RAG-süsteemi abil saab toetada teisigi terminitöö etappe, nt pakkuda mõistekirje koostamiseks välja vajalikku infot, aidata visandada mõistesüsteemi, viidata võimalikele probleemidele algtekstis jm. Süsteem ei suuda terminoloogi küll asendada, ent abivahendina on RAG terminiarenduses perspektiivikas.

***

"Applying artificial intelligence in specialized language development: The example of Estonian national defense terminology"

This article explores the application of Retrieval-Augmented Generation (RAG) for extracting definitions and source references in defense terminology. By integrating external data during generation, RAG enhances large language models, reducing hallucinations and improving accuracy. This capability is critical in fields like defense terminology, where precise definitions and reliable references are essential. The study found that RAG accurately extracts definitions and references from foundational documents, achieving a precision of 89.6% for definitions and 98.9% for references. However, challenges exist, particularly in parsing documents like PDFs, which can lead to occasional inaccuracies. These issues highlight the importance of terminologists for ensuring data quality, as RAG cannot fully replace human oversight. Despite these limitations, RAG holds promise as a tool to support terminology work by efficiently retrieving relevant data and assisting in dictionary compilation. It also offers potential for enhancing other stages of terminology development, such as suggesting synonyms and identifying issues in source texts. Further improvements, such as refining document parsing and reference validation, could enhance RAG’s reliability, making it a valuable tool for terminologists.