TRANSLATING THE UNTRANSLATABLE IN KAZAKH FAIRY TALES: HUMAN vs AI

Abstract

While the roles of human translators and machines have shifted with the development of such engines as Google Translate and DeepL, the issue of translating the untranslatable or culture-specific items from Kazakh into English still remain challenging. The lack of domain-specific training of machine translation tools as well as the underrepresentation of such languages as Kazakh in diverse textual corpora can lead to misinterpretation of the untranslatable, translation errors, and textual incoherence, and translations that might not feel natural to the target reader. To effectively analyze translation inconsistencies, two prompt-based strategies, using GPT-4o, were developed depending on the results of both human and machine translations.

Analyses demonstrate that when an accurate prompt is generated, the current system available for public use makes fewer critical errors. While shedding light on the integration of human and machine translation strategy, this study argues for well-informed translation prompts incorporated with human translation to improve machine translation quality.

Results also show that, though the translation quality is inconsistent, GPT-4o translation with human assistance enhances the emotional and epistemological perspective of the translation of the untranslatable.

The paper will also argue that closest equivalent of the excerpt should be attempted when translating from the source language.

The research is valuable from both theoretical and practical perspectives as it enhances the understanding of the CSIs as it offers insights into how different translation prompts affect cultural transmission. It will also identify limitations and potential improvements for AI translation tools like Google Translate or ChatGPT in the Kazakh language as a lower resource language in Natural Language Processing.