Can Artificial Intelligence Help Bridge the Knowledge Gap? A Contrastive Study between Translations rendered by AI and Human Translation

Abstract

Recent advancements in artificial intelligence (AI) have catalyzed unprecedented transformations across multiple domains, particularly in translation services, through the advent of sophisticated neural language models (NLMs) and large language models (LLMs). The proliferation of machine translation technologies, including NLM-based platforms such as Google Translate, Microsoft Translator, DeepL, and Amazon Translate, has facilitated cost-efficient and expeditious translation processes. Furthermore, computer-aided translation (CAT) frameworks, including Memoq, Trados, and Memsource, have substantially augmented translation methodologies. The recent emergence of advanced LLMs, notably ChatGPT and GPT-4, has demonstrated remarkable progress in contextual comprehension and textual interaction, approximating human-like engagement with written material.

This study investigates the efficacy of Artificial Intelligence, specifically GPT-4, in translating seminal academic works. The research conducts a comparative analysis of three translations of Chapter Three ("Language and interpretation: philosophical reflections and empirical inquiry") from Noam Chomsky's New Horizons in the Study of Language and Mind: (1) an AI-generated translation produced by GPT-4, (2) a translation by three professional translators from Yemen who incorporate GPT-4 into their workflow, and (3) the published Arabic translation by Adnan Hassan, titled "آفاق جديدة في دراسة اللغة والعقل". Through a mixed-methods approach combining automated metrics (BLEU, METEOR, and ROUGE-L) and human evaluation across multiple quality dimensions, this study provides a comprehensive assessment of AI's capabilities and limitations in academic translation.

The findings reveal that while GPT-4 demonstrates impressive capabilities in generating fluent Arabic text and handling general academic language, it struggles with specialized philosophical terminology, cultural adaptation, and preserving nuanced theoretical distinctions. The Human-AI collaborative approach substantially outperforms the GPT-4-only translation across all quality metrics and approaches the quality of the published translation in terms of fluency and readability, while still lagging in terminological consistency and cultural adaptation. Notably, the Yemeni translators reported a 40% reduction in translation time when using GPT-4 as a first-draft tool, suggesting significant potential for increasing translation productivity through AI integration.