سامانه همایش های دانشگاه آزاد اسلامی واحد اصفهان (خوراسگان)

صفحه اصلی / همایش ملی تکنولوژی و تدریس هم افزایی نوین در فرایند یادگیری

Exploring Naturalness in AI-Generated English-to-Persian Translations: Insights from ChatGPT's Performance

نویسندگان :

Mostafa Naemi Kokaneh¹ Mostafa Amiri² Mojgan Kouhpayehzadeh Esfahani³

1- دانشگاه آزاد اسلامی واحد ورامین و پیشوا 2- دانشگاه آزاد اسلامی واحد ورامین و پیشوا 3- دانشگاه آزاد اسلامی واحد ورامین و پیشوا

کلمات کلیدی :

ChatGPT،AI،Machine Translation،Naturalness،Translation Shifts

چکیده :

This study investigated the analytical effectiveness of ChatGPT in rendering natural-sounding translations of descriptive English texts into Persian. Focusing on a 5,000-word excerpt from Erik Ringmar's History of International Relations (2019), the research employed a descriptive analytic design, utilizing Catford's (1965) translation shift model framework for machine translation (MT) quality assessment. The study aimed to identify the types and frequencies of shifts employed by each MT system – structure, class, unit, and intra-system shifts, and to analyze how these shifts impacted the perceived naturalness of the translated texts for Persian-speaking audiences. Data collection involved independent translations of the source text excerpt by ChatGPT, followed by a meticulous analysis of the resulting translations. It must be stated that the researcher himself has found all the content. The findings revealed distinct differences in the translation strategies employed by this AI. ChatGPT exhibited a great propensity for structure and class shifts, indicating a more nuanced approach to adapting the source text to the linguistic and cultural conventions of Persian. This resulted in translations that were perceived as more natural and fluent by Persian speakers, particularly in conveying culturally sensitive concepts and idiomatic expressions. ChatGPT's specific choices within these categories often resulted in more idiomatic phrasing. While the Chi-square analysis did not reveal statistically significant differences in the overall distribution of shift types, the qualitative analysis highlighted the crucial role of these shifts, particularly level and class shifts, in contributing to the perceived naturalness of ChatGPT's translations. The study underscored the importance of moving beyond mere accuracy in MT evaluation and incorporating naturalness as a key criterion, particularly in applications requiring culturally sensitive and nuanced translations.