A Comparative Analysis of Large Language Models for Bilingual Term Extraction in Spanish-Arabic Interpreting and Translation

Abstract

The burgeoning capabilities of Large Language Models (LLMs) are profoundly impacting Natural Language Processing (NLP), with their application in terminology extraction gaining increasing scholarly attention [4]. Terminology extraction is a cornerstone of professional interpreting and translation, indispensable for upholding semantic precision and discursive fluency in specialised communication [1]. Automatic Terminology Extraction (ATE) methodologies endeavour to mitigate the arduous demands of manual terminology management by generating ranked lists of candidate terms from domain-specific corpora [2]. Despite the remarkable capabilities of LLMs, often attributed to their sophisticated training paradigms [4], empirical evaluations of these models for ATE purposes remain notably scarce, particularly concerning linguistically divergent pairs such as Spanish-Arabic. Existing research predominantly focuses on European language combinations, thereby creating a critical lacuna in understanding AI's efficacy in non-Indo-European linguistic contexts. Given the inherent structural and semantic disparities between Spanish and Arabic, a comprehensive assessment of AI's performance in this domain is imperative for ensuring the reliability of LLM-driven tools in professional linguistic workflows [5]. This study undertakes a comparative evaluation of AI tools—specifically ChatGPT, DeepSeek, Gemini, and Manus—for their proficiency in extracting bilingual Spanish-Arabic terminology across two distinct specialised domains: ophthalmology (medical) and tourism. Utilising a methodological framework encompassing Precision, Recall, F-score, and Accuracy [3], this research provides a granular assessment of each tool's capacity for accurate and contextually relevant bilingual term identification. The findings contribute to the theoretical and practical advancements in AI-assisted terminology extraction, offering insights into the evolving landscape of AI integration within interpreting and translation studies.