Translating Cao Zhi's Poetry: Cultural Turn Perspectives on Human and AI Translation

IRASS Journal of Arts, Humanities and Social Sciences https://irasspublisher.com/journal-details/IJAHSS ISSN (Online) 3049-0170 This is an open access article under the CC BY-NC license 58 JOURNAL COVER PAGE Translating Cao Zhi's Poetry: Cultural Turn Perspectives on Human and AI Translation Xiny\’u Zheng* School of Foreign Studies, University of Shanghai for Science and Technology Corresponding Author Xiny\’u Zheng School of Foreign Studies, University of Shanghai for Science and Technology Article History Received: 28 / 10 / 2025 Accepted: 14 / 12 / 2025 Published: 26 / 12 /2025 Abstract: This study examines the English translation of Cao Zhi's poetry from the perspective of the cultural turn, comparing Xu Yuanchong's translations with those generated by the large language models DeepSeek and ChatGPT. Cao Zhi's poetry features concise imagery, rich cultural connotations, and multi-layered emotional expression, making its translation crucial for cross-cultural communication. The study analyzes translations across dimensions such as cultural transmission, imagery transformation, emotional atmosphere, and the realization of cultural function, while also exploring the impact of prompt optimization on the models' translation strategies. Results show that Xu Yuanchong, as a culturally grounded translator, achieves cultural and aesthetic regeneration through creative adaptation. The two language models excel in semantic transmission but adopt different strategies for cultural reconstruction: DeepSeek emphasizes preserving cultural markers and contextual integrity, whereas ChatGPT favors emotional and universalized expression. Prompt guidance can partially influence the models' cultural rendering, poetic structure, and imagery handling, but cannot fully replace human translators' agency in cultural judgment and aesthetic choice. This study contributes to the understanding of poetry translation from a cultural perspective and provides insights for human-AI collaborative translation and model optimization. Keywords: Cao Zhi; poetry; cultural turn; large language models; translation. How to Cite in APA format: Zheng, X. (2025). Translating Cao Zhi's Poetry: Cultural Turn Perspectives on Human and AI Translation.IRASS Journal of Arts, Humanities and Social Sciences, 2(12), 58-63. Introduction The Jian'an period stands as a pivotal era in the history of ancient Chinese literature, marked by an extraordinary emergence of literary talents. Cao Zhi is undoubtedly one of the most accomplished writers. He was acclaimed in Grading of Poets as the ―preeminent figure of the Jian'an era‖, a testament to his lofty status in the literary world of his time. Cao Zhi's poetic works can generally be divided into two phases, demarcated by the ascension of his brother Cao Pi to the throne. His earlier poems often depict the privileged life of a nobleman and convey a spirited ambition to achieve merit and glory. So profoundly shaped by the political climate under his brother Cao Pi, Cao Zhi's later works give poetic expression to a profound inner melancholy, steep political frustration, and the anguish of ideals that remained beyond reach. In the domain of five-character poetry, Cao Zhi achieved particular distinction. He is regarded not only as a master of the fivecharacter verse but also as a versatile literary talent. Liu Xie, in The Literary Mind and the Carving of Dragons·Appraisal of Talents, praised him as ―Zijian, swift in thought and brimming with talent, whose poetry is elegant and whose prose is transcendent.‖ Cao Zhi exerted a profound influence on the evolution of the fivecharacter poetic form and left an indelible imprint on the Chinese literary tradition. He is widely considered one of the most influential poets in the history of Chinese literature. In the field of translation studies, advances in artificial intelligence have brought unprecedented convenience to crosslinguistic communication. AI-powered tools, particularly large language models, are capable of generating translations on a large scale within short timeframes, significantly improving efficiency and reducing costs. This opens up new possibilities for translation practice. However, notable limitations persist when AI is applied to classical Chinese poetry. Such poetry is characterized by mutable imagery, extreme lyrical conciseness, and deep cultural allusions embedded within a rich aesthetic tradition. Translating these works demands considerable cultural literacy and aesthetic sensitivity— qualities that current AI systems have yet to fully replicate. Therefore, the role of human translators remains irreplaceable in the English translation of classical Chinese poetry. In recent years, research has increasingly focused on how human translators can effectively convey the cultural and artistic essence of poetic works. Still, studies exploring the application of large language models to the translation of classical poetry remain relatively scarce. Against this backdrop, this paper selects Cao Zhi's poetry as its subject, conducting a comparative analysis of translations produced by large language models and those by human translators. The aim is to offer insights that may benefit the cross-cultural dissemination of classical Chinese poetry. This study takes Cao Zhi’s poetry as its primary corpus and conducts a comparative analysis of translations produced by the human translator Xu Yuanchong and by DeepSeek and ChatGPT. It aims to examine how different translations perform in terms of cultural transmission, imagery transformation, and poetic reconstruction, thereby providing insights for the cross-cultural dissemination of classical Chinese poetry and the application of AI in literary translation. Accordingly, this study addresses the following research questions: 1. How do Xu Yuanchong’s translations and those generated by DeepSeek and ChatGPT differ in their representation of cultural connotations, imagery systems, IRASS Journal of Arts, Humanities and Social Sciences Vol-2, Iss-12 (December-2025): 58-63 Vol-2, Iss-12 (December-2025) 59 and affective atmospheres in Cao Zhi’s poetry? In what ways does each translation reconstruct the cultural functions of the original text? 2. What distinct strategies do the human translator and large language models adopt in the English translation of classical Chinese poetry? What are the salient features and limitations of LLMs in conveying cultural information, transforming imagery, and reproducing poetic qualities? 3. How do culturally oriented prompts influence LLMs’ translation strategies, cultural explicitation, and poetic construction? In what ways is the regulatory function of prompts manifested in the translation behavior of large language models? The Theory of Cultural turn Bassnett, as one of the representatives of the cultural school, holds that translations should be equivalent, and that equivalence is not confined merely to the linguistic level. The purpose of translation lies in achieving a cultural transfer; the true aim of translation should be to meet the needs of the target culture or a specific target group. In the translation process, translators should therefore consider contextual, functional, and cultural equivalence. Bassnett endorses Catford's classification of two types of untranslatability—linguistic untranslatability and cultural untranslatability. The former refers to cases where the sourcelanguage structure is absent in the target language, making it impossible to find an equivalent expression, which arises from structural differences between the source and target languages. The latter is more challenging and primarily occurs when the target language cannot reproduce the contextual characteristics of the source text, particularly symbolic metaphors with cultural significance. This constitutes cultural untranslatability. For example, the imagery and cultural allusions in classical Chinese poetry often carry rich cultural connotations, which are difficult to fully convey to target-language readers in translation. In Translation, History and Culture, co-edited by Bassnett and Lefevere, the concept of the ―cultural turn‖ in translation is explicitly proposed. The cultural turn emphasizes the cultural dimension of translation, viewing translation not only as an equivalence-based linguistic transfer but also as a medium for cultural exchange and transfer. This perspective opens new horizons for translation studies, shifting the focus from purely linguistic equivalence to cultural dimensions, and highlights the cross-cultural functions and impacts of translation. The core of the cultural turn includes the cultural functions of translation, cultural identity and power relations, as well as rewriting and reception. Translation is a form of ―rewriting,‖ where translators adapt the source text to meet the requirements of the target culture. Such adaptations have a significant impact on the reception of the translation and on cultural relationships. Through the lens of the cultural turn, translation studies are no longer restricted to linguistic equivalence but instead focus on cross-cultural communication and influence, providing new guidance for translation practice and assisting translators in addressing cultural differences, thereby enabling them to integrate the cultural function of translation into the process of cultural dissemination. Methodology Research Materials and Data Sources This study takes the classical poetry of Cao Zhi, a representative poet of the Jian’an period, as its primary research material, with the aim of examining how his works are rendered in cross-cultural transmission in terms of cultural representation and poetic reconstruction. To systematically compare the translation strategies and cultural transmission effectiveness of human translators and large language models, the study collects the following types of English translations: (1) Human translation: Xu Yuanchong’s English versions, selected for their combination of cultural fidelity and poetic creativity. (2) LLM-generated translations: Translations produced by two large language models, DeepSeek and ChatGPT. (3) Prompt-based LLM translations: For each model, two categories of prompts were designed to generate translations. Basic prompt: A simple instruction to translate, without additional context, serving as the baseline. Cultureoriented prompt: Prompts emphasizing historical context, cultural background, and the reconstruction of poetic imagery. All poems and their translations were compiled into text-based datasets to support systematic analysis. Textual Analysis Methods This study primarily employs qualitative comparative textual analysis within the framework of cultural translation theory. The specific methods include: (1) Comparative textual analysis: Corresponding lines and segments across the three types of translations are examined with respect to cultural transmission, imagery transformation, affective and aesthetic rendering, and the reconstruction of cultural functions. (2) Prompt-effect analysis: By comparing translations generated under the basic prompt and the culture-oriented prompt, the study investigates how different prompts influence LLM translation strategies, cultural explicitation, and the construction of poetic qualities. Results and Discussion In literary translation, the issues of untranslatability and the cultural shift focus on how cultural elements within translation influence language and translational practice. This is particularly relevant for translating works rich in cultural meaning and imagery, such as classical Chinese poetry. These concepts not only reflect the permeation of culture into language but also reveal, through translation, the complex relationship between language and culture—manifesting goals such as cultural adaptability and functional equivalence in translation. Cao Zhi's Written While Taking Seven Paces(《七步诗》), a recognized classic of classical Chinese poetry, serves as the object of this analysis. This section compares English translations of the poem by Xu Yuanchong, DeepSeek, and ChatGPT, examining aspects such as the transmission of cultural background, the conversion of imagery, the conveyance of emotional resonance, and the realization of cultural function. Through this comparison, the study illustrates how different translation strategies shape the representation of cultural meaning and the achievement of intended cultural and functional purposes. Example 1 《七步诗》 煮豆持作羹，漉豉以为汁。 萁在釜下燃，豆在釜中泣。 本自同根生，相煎何太急？ IRASS Journal of Arts, Humanities and Social Sciences Vol-2, Iss-12 (December-2025): 58-63 Vol-2, Iss-12 (December-2025) 60 Xu Yuanchong's Translation:  Written While Taking Seven Paces  Pods burned to cook peas，  Peas weep in the pot：  Grown from the same trees，  Why boil us so hot? DeepSeek's Translation:  The Poem of Seven Steps  Beans boiled to make a broth fine,  Ground slow to draw their bitter juice.  The stalks burn fierce beneath the pot,  While beans above cry out their truth:  "We share the selfsame root and frame,  Why burn me with such ruthless flame?" ChatGPT's Translation:  The Quatrain of Seven Steps  Boiling beans to make a soup,  Filtering the beans to extract the broth.  Beanstalks burn beneath the pot,  While the beans inside weep.  Born of the same root,  Why should you torment each other so harshly? This poem was composed under duress: Emperor Cao Pi, threatening Cao Zhi with execution, commanded him to produce a poem within seven paces. In response, Cao Zhi employed the metaphor of ―beans and beanstalks‖ to allegorize fraternal conflict. In terms of cultural background transmission, ―豆萁‖ represents a culturally specific symbol rooted in China's agrarian society, lacking a direct equivalent in Anglo-American culture. The term ― 羹‖ refers to a paste-like or gelatinous food prepared by steaming or boiling. DeepSeek's rendering as ―broth fine‖ approaches the original meaning of ―羹‖; ―broth‖ denotes a boiled soup, while the modifier ―fine‖ subtly underscores the irony of refinement amid cruelty, thereby facilitating cultural transmission. By contrast, ChatGPT generalizes ―羹‖ as ―soup,‖ stripping the term of its cultural connotations and reducing it to ordinary culinary language. Regarding the translation of objects and metonymy, the ―豆 -萁-釜‖ triad presents considerable difficulty in English. The beanstalk, grown from the same root as the beans, becomes both agent and victim of harm, while the ―釜‖ (an ancient cauldron) symbolizes a confined space of power. The Chinese phrase ―同根 生‖ carries the collective unconscious of China's agrarian civilization—it denotes not only a shared biological origin but also an interconnected life sustenance and ethical bond. The English phrase ―the selfsame root,‖ in the target-language context, merely indicates common origin, losing the profound sense of a life community bound together yet trapped in self-inflicted harm. DeepSeek's translation of ―同根生‖ as ―the selfsame root and frame‖ retains the denotation of common origin through ―root,‖ while the addition of ―frame‖ explicitly conveys the implicitly shared ―life community‖ in the Chinese—referring both to the symbiotic physiological structure of plants and metaphorically to the fabric of human relations. This approach mitigates potential misunderstanding from using ―root‖ alone, enabling English readers to apprehend the imagery on both biological and social levels. Moreover, the expression ―ruthless flame‖ personifies the natural element of fire, attributing moral agency to it and thereby intensifying the poem's ethical accusation. In translating the imagery, the DeepSeek version progressively strips away layers of Chinese agrarian cultural knowledge, yet uses this gradient of imagery to re-convey the original poem's political allegory and emotional depth to targetlanguage readers. Xu Yuanchong's translation uses ―burned‖ to convey the violence of persecution and constructs a botanical connection through ―peas‖ and ―pods‖ to preserve the ―peas‖ imagery within the cultural context of ―same-origin strife.‖ The term ―pot‖ serves as a generic container, helping target readers comprehend the enclosed spatial consciousness in the original without introducing cultural barriers. The word ―same‖ extends the Chinese notion of ―同根生‖ to the English cultural concept of ―family tree,‖ reconstructing a cognitive model recognizable to the target audience from Chinese agrarian culture. The ChatGPT translation's use of ―torment each other‖ entirely erases the unidirectional persecution, violence, and power dynamics present in the original, reducing Cao Pi's political oppression of Cao Zhi to mutual conflict. This approach reflects a domesticating strategy that weakens the cultural and political allegorical force of the source text. In terms of emotional and aesthetic representation, the DeepSeek version renders the含蓄委婉 ―釜中泣‖ as the direct ―cry out.‖ While this achieves equivalent emotional intensity, it significantly diminishes the multifaceted emotional expression of the original. The uniquely repressed affective mode of ―泣（weep ）,‖ which in Chinese implies audible but tearless weeping, is lost when converted into ―cry out.‖ This not only strips away the sense of spiritual oppression conveyed by the enclosed vessel (―釜‖) but also transforms the characteristically restrained ―哀而不伤‖ emotional aesthetic of Chinese literati into a Western-style dramatic emotional outburst. Xu's translation, through ―Pods burned / Peas weep,‖ emphasizes rhythmic regularity and unity, enhancing emotional impact through sonorous phrasing. By establishing a progression of heat through ―burned – boil – hot,‖ it transforms the implicit ―煎‖ in the Chinese into a physically perceptible pain for the English reader. This synesthetic effect avoids the cultural depreciation that might result from a direct translation of ―泣,‖ while using tactile metaphor to achieve emotional resonance. Additionally, by shifting from the original third-person narrative to direct speech—―why boil us so hot‖— Xu's version skillfully negotiates the differences between Chinese and Western lyrical traditions. The ―weeping beans‖ in the Chinese poem represent an indirect expression of the literati's practice of using objects to convey ideals; by giving the beans a voice, Xu's translation preserves the Eastern philosophical perspective of ―unity of thing and self,‖ while aligning with the Western lyrical tradition that emphasizes subjective expression. The ChatGPT translation employs technical terms such as ―filtering…extract‖ to interpret the cooking process, reducing the politically charged act of ―漉豉‖ to a purely technical operation. This entirely dissolves the emotional violence embedded in the dietary imagery of the original. At the cultural level, Xu Yuanchong’s translation demonstrates a clear capacity for interpretive refinement: by distilling imagery and purifying emotion, his version effectively ―rewrites‖ the poem’s implicit admonition about political persecution in a form more resonant with the target culture. IRASS Journal of Arts, Humanities and Social Sciences Vol-2, Iss-12 (December-2025): 58-63 Vol-2, Iss-12 (December-2025) 61 DeepSeek’s translation, through creative transformation, manages to bridge the cultural gap between source and target traditions, preserving the poem’s cultural depth while crafting a poetic register compatible with English literary norms. In contrast, ChatGPT’s version shows limited ability to enact such culturally motivated rewriting, revealing a lack of deeper cultural engagement. These findings collectively reaffirm Bassnett’s view of translation as a process of cultural rewriting and concretely illustrate how different translation strategies vary in their effectiveness when handling cultural meanings and accommodating the expectations of the target culture. The following excerpt selects two typical and important lines from Cao Zhi's Seven Sorrows (《七哀诗》). In the first line, the poet uses the rhetorical device of ―比‖ (comparison) to liken himself to a grieving wife, expressing a willingness to transform into the southwest wind (西南风, southwest wind) in order to return to his lord's embrace. Cao Zhi longs for familial reconciliation and hopes to serve alongside his elder brother Cao Pi and nephew Cao Rui. Hence, he states his wish to become a southwest wind and reenter their embrace. The second line employs ―赋‖ (expository description) to convey the experience and sentiment of a neglected wife, implicitly reflecting his own sense of estrangement and frustration caused by his elder brother's alienation. Example 2 七哀诗 (excerpt) 愿为西南风，长逝入君怀。 君怀良不开，贱妾当何依？ Xu Yuanchong Translation:  Seven Sorrows  Would that I were the wind from the southwest.  That I could rush across the land to your breast!  From your embrace, if you should shut me out,  Where should I go? Where should I roam about? DeepSeek Translation  Seven Sorrows  Could I become the southwest breeze,  To rush into your arms with ease?  But your arms won't open wide—  Where can this humble wife abide? ChatGPT Translation  Seven Sorrows Poem  I wish to be the southwest wind,  To pass forever into your embrace.  But your heart remains tightly closed;  Where then can my lowly wife turn for refuge? In terms of cultural transmission, the term ―西南风‖ (southwest wind) carries emotional and symbolic connotations in the original poem. All three translations retain the image, but Western readers may not fully grasp its cultural significance. Translators can clarify the symbolic meaning of ―西南风‖ through footnotes or commentary. Xu Yuanchong's use of ―my lord‖ to translate ―君‖ is problematic. In Chinese tradition, ―君‖ is a multilayered concept: in a marital context, it can refer to ―husband‖; it also carries connotations of hierarchical authority (― 君为臣纲‖) and, in political terms, hints at the lord-subject relationship between Cao Zhi and Cao Pi. The English term ―lord,‖ derived from European feudal systems, cannot convey the Confucian ethical subtleties embedded in ―君,‖ thereby flattening the complex interplay of social relationships and emotional tension in the original text. In contrast, the other three translations use the second-person pronoun ―you,‖ preserving the ambiguity of ―君,‖ which can signify both marital and political references, allowing English readers interpretive space while maintaining the speaker's feelings of grievance and longing. Regarding emotional expression, Xu's ―Would that I were‖ (subjunctive) and DeepSeek's ―Could I‖ (interrogative) convey stronger emotional nuance than ChatGPT's declarative phrasing. Xu's ―Would that I were‖ is formal and literary, expressing a strong yet unattainable desire to traverse mountains and rivers to reach the beloved, imbued with resignation and sorrow. His depiction of the wind rushing across the land creates tension between aspiration and the vast, unattainable distance. The subsequent ―shut me out‖ interrupts this emotional build-up, ending with the drifting ―roving about,‖ reflecting the complex predicament of a woman constrained by Confucian norms. DeepSeek's use of ―breeze‖ rather than ―wind‖ conveys the subtle, restrained Chinese aesthetic of grief without complaint, whereas ChatGPT's translation is more mechanical, with discontinuous emotional expression. From the above analysis, in terms of cultural function, Xu Yuanchong and DeepSeek more fully convey the Chinese cultural imagery, with Xu prioritizing emotional and phonetic beauty, whereas ChatGPT achieves surface semantic correspondence but weakens cultural transmission. 5. The Modifying Effect of Prompt Input on the Quality of Model-Generated Translations With the deepening application of LLMs in the field of translation, prompts have gradually become a key factor influencing the quality of model outputs. Classical poetry, with its highly distinctive formal, cultural, and aesthetic features, poses particular challenges for LLMs. Without explicit translation instructions, LLMs often struggle to capture the deeper semantic structures and aesthetic mechanisms of the poems, making it difficult to generate translations with both coherence and cultural depth. Building on the comparative analysis presented in previous chapters, this section further investigates whether prompt optimization can enhance the cultural representation and aesthetic expression of LLMs in the English translation of Cao Zhi's poetry. Prompt Types and Experimental Method To examine the influence of prompts on translation, this study designed four representative types of prompts for the DeepSeek and ChatGPT models and conducted controlled tests. (1) Basic Prompt: Simply instructing ―Translate the following poem into English,‖ as in Examples 1 and 2, served as the control group, allowing observation of the models' natural translation tendencies without additional guidance. (2) Culture-oriented Prompt: Emphasizing the reconstruction of historical context, cultural background, and classical references, the input was: ―You are a poet. Translate the following poem into English, and appropriately present the poem's historical background and cultural meaning in the translation.‖ IRASS Journal of Arts, Humanities and Social Sciences Vol-2, Iss-12 (December-2025): 58-63 Vol-2, Iss-12 (December-2025) 62 Analysis of Prompt Effects on Translation Quality Translation Strategies under Basic Prompt Under the basic prompt, the models tended to rely on literal translation, often handling cultural information such as allusions and symbolic imagery inadequately. This tendency reflects the models' general translation mode, which prioritizes semantic equivalence in the absence of explicit task instructions. This section uses the basic prompt as a reference to examine how culture-oriented and aesthetic-oriented prompts influence translation strategies, textual style, and cultural realization. Culture-oriented Prompt: Cultural Presentation and Context Reconstruction When guided by the culture-oriented prompt, both models exhibited a marked tendency toward ―cultural context visualization,‖ enhancing narrative relationships, identity references, and emotional hierarchies in their translations, thereby attempting to reconstruct the cultural framework of classical poetry. The following translations were produced for an excerpt of Seven Sorrows (《七哀诗》) under a culture-oriented prompt: DeepSeek Translation:  I long to be the southwestern breeze,  That drifts unseen into your embrace.  Should your heart bar its gates to me,  On whom, my lord, shall this lowly one rely? ChatGPT Translation:  I wish to be the southern breeze,  to drift forever into your embrace.  Yet your heart remains closed to me—  what refuge then for one so lowly and forlorn? DeepSeek emphasizes hierarchical relationships present in the classical context; for instance, the phrase ―my lord‖ in the last line reproduces the original poem's lord-subject reference. ChatGPT, in contrast, maintains a more abstract emotional focus, rendering humility as ―one so lowly and forlorn,‖ which reduces the political-hierarchical nuance while strengthening the universality of individual emotion. Both models employ expressions such as ―embrace‖ and ―bar its gates‖ to construct more concrete contexts, giving imagery such as ―wind‖ (风) and ―embrace‖ (怀) explicit cultural metaphorical function. DeepSeek's ―drifts unseen‖ reinforces the symbolic notion of the wind as a stealthy presence in the original poem, whereas ChatGPT's ―to drift forever‖ emphasizes the subject's longing for enduring proximity, demonstrating different expansions of imagery. Regarding cultural context and allusions, DeepSeek tends to cultivate classical flavor through register, titles, and syntactic style, whereas ChatGPT emphasizes emotional fluidity and readability in modern English poetic conventions. The culture-oriented prompt effectively guides both models to highlight cultural information, strengthen cultural cues, and reconstruct contextual relationships, resulting in translations that better reflect the cultural awareness of the original text compared to the basic prompt. However, their strategies differ: DeepSeek focuses more on recreating the structural and aesthetic qualities of classical context, while ChatGPT leans toward reinterpreting emotional and cultural connotations through modern English expression. These findings suggest that LLMs demonstrate high controllability in classical poetry translation, and prompts are key to achieving such control. Future research can further explore how fine-grained prompt design affects different layers of poetry translation, offering new methods and possibilities for poetry translation in the age of artificial intelligence. Conclusion Based on the theoretical framework of the cultural turn, this study conducted a comparative analysis of English translations of Cao Zhi's poetry by Xu Yuanchong, DeepSeek, and ChatGPT. Through systematic examination of cultural context reconstruction, imagery conversion, emotional resonance conveyance, and cultural function realization, the analysis reveals distinct characteristics and emphases between human and LLMs, while also demonstrating the significant influence of different prompt designs on model output. Xu Yuanchong’s translations demonstrate a pronounced cultural awareness and refined aesthetic judgment. Drawing on deep cultural knowledge and extensive poetic experience, he is able to creatively reconstruct the poem’s imagery system, cultural allusions, and aesthetic intent, producing translations that retain the cultural depth of the original while exhibiting an independent poetic quality. By contrast, large language models in their default state rely primarily on statistical patterns and corpus-based language generation, making it difficult for them to spontaneously grasp the intricate cultural layers and symbolic structures of classical Chinese poetry. As a result, their output tends to remain at the level of semantic transfer. However, when guided by culturally oriented or role-based prompts, these models show improved cultural explicitation and greater precision in expression, and their overall translational orientation becomes more controllable, revealing a notable degree of malleability. In terms of imagery and poetic quality, DeepSeek generally produces more stable renderings, whereas ChatGPT is more prone to semantic paraphrasing or structural reorganization; nonetheless, both models display heightened cultural sensitivity when provided with explicit instructions. Overall, the strength of the human translator lies in the autonomy of cultural interpretation, aesthetic judgment, and expressive strategy, while the potential of large language models depends closely on the design and specificity of the prompts they receive. This study underscores the pivotal role of prompts in shaping model behavior in poetry translation and demonstrates that, through effective prompt engineering, AI systems can significantly enhance their ability to represent cultural meaning and construct poetic effect in the translation of classical Chinese poetry. Although this study provides a systematic analysis of English translations of Cao Zhi's poetry from the perspective of the cultural turn, several limitations remain. The relatively small number of poems selected cannot fully represent LLM performance across different text types. The dynamic updating of LLMs also affects research stability, as version iterations of DeepSeek and ChatGPT may lend transitional characteristics to the findings. Furthermore, despite employing multidimensional cultural analysis methods, the study lacks empirical data on translation reception and target reader responses, indicating the need for further development of the evaluation framework. These limitations suggest directions for future research. Subsequent studies could expand the poetic corpus, optimize prompting strategies, focus on cross-cultural communication effects, explore IRASS Journal of Arts, Humanities and Social Sciences Vol-2, Iss-12 (December-2025): 58-63 Vol-2, Iss-12 (December-2025) 63 human-AI collaborative translation, and dynamically track translation strategy evolution. Such efforts would contribute to a more comprehensive understanding of classical poetry translation patterns and cultural transmission mechanisms, while providing new insights into the interaction between human translation and intelligent translation assistance. References 1. Chen, T. (2015). On the new cultural perspective in translation studies: Exploring the essence of the ―cultural turn‖ in Western translation studies. Hunan Social Sciences, (01), 206–209. 2. Appreciation of Poetry from Pre-Qin, Han, Wei, and Six Dynasties (Vol. 2). (2023). Shanghai: Shanghai Ancient Books Publishing House. p. 463. 3. Duan, F. (2020). A study on the subjectivity of the translator in literary translation from a cultural perspective (2nd ed.). Chengdu: Sichuan University Press. p. 106. 4. Liu, J. P. (2022). Chinese writing and general knowledge (2nd ed.). Wuhan: Wuhan University Press. pp. 275–276. 5. Ma, L. L. (2021). A study on Jian’an fu. Shanghai: SDX Joint Publishing Company. pp. 92–93. 6. Wang, Y. H. (1991). A self-study guide to ancient Chinese literature. Shenyang: Liaoning University Press. pp. 260–261.