Strengths and Limitations of Artificial Intelligence in Conveying Gender Vagueness in Literary Translation

Abstract

Over the past few decades, the field of translation studies has undergone significant technological evolution, progressing from computer-assisted translation (CAT) tools to neural machine translation (NMT) systems and, most recently, to large language model (LLM)-based chatbots. The latter, while not originally designed for translation tasks, demonstrate high efficacy in natural language processing and function as cognitive communication systems (dialogic artificial intelligence (AI) agents). When utilized by translators, such tools can significantly enhance productivity and output quality due to prompt engineering techniques. This study examines the ability of AI systems to process and render in translation the semantics of blurred boundaries, i.e. instances where literary authors create situations of interpretative choice that invite readers’ active meaning construction. The paper evaluates how current AI solutions can: 1) detect intentionally vague elements in a literary text and 2) render them across languages while preserving the author’s intent. Data collection was carried out using comparative analysis of translation results performed by 15 NMT systems (DeepL Translator, Google Translate, Yandex Translate, SYSTRAN, HIX, etc.) and 4 AI chatbots (ChatGPT-4o, Google Gemini, Microsoft Copilot, and Sider). The findings demonstrate that while conventional NMT systems struggle with conveying gender neutrality – a manifestation of semantic vagueness – LLM-based chatbots excel through their generative architecture. Their interactive cognition enables contextual adaptation and alternative phrasing, effectively bridging the “gender vagueness gap” in translation.