Jul 2026· Journal of Translation and Language Studies· 0 citations· 20 references
TL;DR
The study finds that each prompt orientation produces distinct and observable shifts in diction, imagery construction and formal expression, and re-conceptualizes prompts not only as technical input instructions, but as purposive regulators of translation, offering a translation-theoretic framework for analysing LLM behaviour and practical guidance for designing Skopos-informed prompts for AI-assisted literary translation.
Abstract
Large language models (LLMs) have rapidly expanded the horizon of AI-assisted literary translation, yet the mechanisms of how prompts construct translation purpose are theoretically underexplored. This study, based on Vermeer’s Skopos Theory, conceptualizes prompts as a “digital translation brief” and examines the systematic regulation of four function-oriented prompts: baseline (P0), reader-oriented (P1), form-oriented (P2), and culture-oriented (P3) in DeepSeek’s translation strategies for English-to-Indonesian poetry translation, with Emily Dickinson’s Hope is the Thing with Feathers as the source text. The study finds, through comparative close reading across the four prompt conditions, as well as against two published human translations, that each prompt orientation produces distinct and observable shifts in diction, imagery construction and formal expression: P1 sacrifices collocational stability for literary register and emotional intensity at the expense of; P2 replicates structure but risks over-compliance which undermines target-reader adequacy; P3 allows discourse-level metaphorical coordination and selective dependence on the source-text. Human translators’ choices emerge from cultural memory, aesthetic intentionality, and poetic agency. DeepSeek’s functional adaptability is largely a matter of probabilistic generation and prompt compliance. The findings re-conceptualize prompts not only as technical input instructions, but as purposive regulators of translation, offering a translation-theoretic framework for analysing LLM behaviour and practical guidance for designing Skopos-informed prompts for AI-assisted literary translation.
The paper argues that NLP should operate as an interpretive assistant rather than an autonomous literary translator in translating Iraqi poetry into English, and proposes a culturally aware, human-in-the-loop framework for supporting literary translation.
Whaj Mneer Esmail· Iraqi Literary and Cultural...· 0 citations
The results show that LLMs and Google Translate consistently outperform specialized MT systems in terms of fluency, meaning preservation, and lexical-thematic alignment.
Beatriz Ribeiro Borges, P. H. R. Gabriel, E. Faria· International Journal of Dat...· 0 citations
: In the era of digital transformation and the rapid advancement of generative artificial intelligence, the translation of idiomatic expressions has become a crucial benchmark for evaluating the cognitive and linguistic capabilities of Large Language Models (LLMs). This paper presents a detailed analysis of research conducted on a corpus of ten English body part idioms taken from the Pioneer B2 textbook used at Singidunum University. The aim of the research was to compare translations generated by the ChatGPT model with solutions from official idiomatic dictionaries, utilising Pavol Kvetko's classification and Mona Baker’s equivalence strategies as the theoretical framework. The analysis encompasses idioms of varying degrees of transparency, ranging from completely opaque to semi-idioms. The study results indicate a 90% accuracy rate in conveying meaning, alongside an unexpectedly high 60% correspondence of keywords in both languages. The research confirms that ChatGPT successfully identifies functional equivalents in the Serbian language, often prioritising the naturalness of expressions over literal translation. This work contributes to the discussion on the role of AI tools as assistants in translation and education, emphasising that while AI shows exceptional dexterity in mapping conceptual fields, human oversight remains essential for the final validation of stylistic nuances. The findings have significant applications for international scientific research, particularly in the domain of applying information technology in foreign language teaching.
Jelena Janackovic, Jovana Bošković, Jelena Mladenović· SINTEZA· 0 citations
Backround - The rapid advancement of Artificial Intelligence (AI) in translation studies has transformed how literary texts are processed, shifting from literal word-for-word transfer to more contextually nuanced approaches. Literary fables demand particular sensitivity to personification, onomatopoeia, and moral messaging.
Urgency of Research - Despite the proliferation of AI translation tools, previous studies have predominantly focused on macro-level quality evaluation (e.g., fluency and accuracy) rather than examining the micro-linguistic strategies AI employs. There remains a significant gap in understanding how different AI models identify and apply established translation techniques within Molina and Albir's comprehensive framework of 18 translation techniques.
Research Objectives - This study aims to evaluate and compare the performance of four AI tools—ChatGPT, Gemini, Claude, and DeepL—in identifying and applying translation techniques in the literary fable "The Clever Rabbit," specifically examining how each model utilizes Molina and Albir's 18 translation techniques to achieve dynamic equivalence.
Research Method - Adopting a qualitative descriptive approach, this study employs purposive sampling to select translation units demonstrating specific techniques. Data were collected through comparative textual analysis of one English source text and four Indonesian target texts, validated through theoretical triangulation and source triangulation.
Research Findings - The findings reveal a clear strategic polarization: generative AI models (ChatGPT, Gemini, Claude) demonstrate dominance in complex transformation techniques such as Modulation (20-23%), Equivalence (12-13%), and Explicitation, reflecting deeper contextual understanding. In contrast, DeepL shows extreme reliance on Literal Translation (>65%) with minimal cultural or stylistic adaptation. ChatGPT excels in local adaptation through generalization and particularization; Gemini stands out in narrative vitality through expressive lexical variation; and Claude offers structural efficiency through precise grammatical reduction.
Research Conclusion & Novelty - This study concludes that while all AI tools can transfer denotative meaning, generative models (LLMs) are superior in applying high-level translation techniques necessary for maintaining emotional nuance, discourse cohesion, and literary appeal. The novelty lies in its micro-linguistic analysis using Molina and Albir's comprehensive taxonomy across four distinct AI platforms, providing unprecedented insight into the "black box" of AI translation strategies. The findings offer practical guidance for educators, researchers, and translators in selecting appropriate AI tools, emphasizing that critical human post-editing remains indispensable for achieving true literary equivalence.
As AI has been increasingly applied in translation, Large Language Models (LLMs) are becoming important tools in translation practice. Both human translators and AI models employ strategies in the translation process, yet comparative studies remain limited. Using a self-developed Translation Strategy Evaluation Scale (TSES), this study drew on scientific, political-economic, and literary texts to compare human and AI translation strategies. Think-Aloud Protocols (TAPs) and AI reasoning outputs were used to examine differences in strategy types and decision-making between two human translators and two LLMs (GPT-5.3 and DeepSeek-V3.2); the Many-Facet Rasch Model (MFRM) was then applied to evaluate strategy use across four dimensions: necessity, compatibility, effectiveness, and consistency. Results show that while both shared 7 basic categories, they diverged in 4 respects. AI used abstraction and nominalization for stylistic formality; human translators performed grammatical monitoring, redundancy reduction, and discourse reorganization. In decision-making, AI followed systematic planning and rule-driven execution, while human translators relied on experiential judgment and reader awareness with ongoing self-monitoring. ChatGPT demonstrated the highest quality of strategy use across all text types. The study sheds light on the process-oriented nature of translation strategies and offers empirical evidence for integrating technological tools into translation pedagogy.
Xuefeng Wu, Chenchen Liu· English Language Teaching· 0 citations
Artificial-intelligence-assisted translation of culturally marked phraseology may produce semantically plausible output while failing to preserve affective valence, pragmatic force, register, or cultural symbolism. This study examines English and Azerbaijani zoonym-based phraseological units in bidirectional AI-assisted translation. A qualitative corpus-based comparative design was applied to 100 units (50 English and 50 Azerbaijani) selected from phraseological dictionaries and literary and digital sources. Translations were generated with ChatGPT (OpenAI GPT-5.5) through the official web interface between 15 and 20 July 2026. Each source item was tested three times in newly initiated sessions using a standardized prompt. Outputs were compared with reference equivalents identified in the cited lexicographic sources and verified by the author across six dimensions: semantic adequacy, idiomatic naturalness, affective equivalence, pragmatic function, cultural appropriateness, and register preservation. The qualitative case analyses illustrate successful preservation of conventional target-language equivalents in some items and literal rendering, metaphorical-image mismatch, reduced emotional expressiveness, pragmatic weakening, or loss of cultural symbolism in others. Because corpus-level frequencies and the complete item-level record are not presented in this version, the findings should be interpreted as qualitative patterns rather than statistical estimates. The study demonstrates the value of context-sensitive, linguoculturally informed evaluation of AI-assisted phraseological translation.