DECODING EGYPTIAN DIALECT WITH GEMINI AI AND GOGOLE TRNSLATE

Abstract

The transformative power of Artificial intelligence (AI) is reshaping industries, and translation is no exception. Colloquial Egyptian Arabic (CEA), a vibrant yet underrepresented dialect in machine translation (MT), provides a unique challenge. This study investigated the potential of the advanced language model Gemini Advanced (GA) to translate CEA into English. Using a corpus of sentences from a popular Egyptian film, GA's performance was compared against Google Translate (GT). Analysis employed communicative translation principles and tailored translation briefs. The results showed that GT consistently produced literal, unidiomatic translations. In contrast, GA demonstrated greater potential for communicative translation, with nearly half of its initial translations accurately conveying meaning. Providing additional context significantly improved GA’s performance, exceeding 80% accuracy in some specific categories, though inconsistencies remained. This research highlights the challenges of translating CEA with models primarily trained on Modern Standard Arabic (MSA) and underscores the potential of AI to address this gap. Successfully decoding CEA could revolutionize our ability to navigate the complexities of this living dialect, with significant applications in subtitling and other digital content domains. While not currently ideal for standalone CEA translation, GA shows significant promise as a powerful tool within a human-AI collaborative workflow.

Date of Award	2024
Original language	American English
Awarding Institution	HBKU College of Humanities and Social Science