Literary Translation Is More Than Replacing Words
Literary translation involves more than transferring information from one language to another. Novels, short stories, and other literary works often rely on metaphors, emotional tones, narrative voices, idiomatic expressions, symbols, and culturally specific references.
Generative AI and large language models have significantly expanded the ability of machines to produce fluent, contextual translations. However, linguistic fluency does not automatically mean that the original meaning, aesthetic character, or cultural sensitivity has been preserved.
For readers, this distinction can be important. A literal or highly standardized translation may communicate what happened in a story but change how the author intended readers to experience it. In literary works, ambiguity, imagery, and cultural references can be part of the meaning itself.
Tambunsaribu's research addresses this issue by examining literary translation through three interconnected dimensions: meaning, style, and cultural representation. Rather than treating these dimensions as separate features, the study examines how a single translation decision can affect all three simultaneously.
Three Human Translators Compared With One AI Model
The research used a qualitative comparative design. The researcher selected portions of English literary texts containing metaphors, idioms, narrative styles, and cultural elements.
The same source material was translated by three human translators and one AI model. Two translation experts subsequently reviewed the interpretation of the results.
Instead of using questionnaires or statistical measurements, the researcher compared the texts directly. The analysis examined whether each translation retained contextual meaning, metaphors, figurative language, tone, narrative voice, idioms, cultural terms, and social practices.
The research process involved documenting the original texts and translations, identifying differences and similarities, grouping the findings into themes, and having two translation experts review the resulting interpretations.
AI Tends to Explain Meaning More Directly
The first major difference appeared when literary expressions carried implicit or contextual meaning.
For example, the sentence “She carried the weight of the whole family on her shoulders” uses a metaphor to describe responsibility. Human translations in the study generally retained the imagery of weight and shoulders. The AI version communicated the basic idea of being responsible for the family but removed the metaphorical image.
A similar pattern appeared in the sentence “He smiled, but there was winter in his eyes.” Human translators tended to preserve the image of winter, while the AI translation interpreted the expression more directly as sadness and coldness.
The AI versions therefore remained understandable, but they gave readers less room to interpret the metaphor themselves. According to the study, human translators were more likely to preserve implicit and contextual meanings, while AI more frequently made those meanings explicit.
Human Translators Preserve Literary Imagery More Consistently
The difference became even more visible in metaphors and figurative language.
In the expression “The night swallowed the village whole,” the human translators retained the figurative idea of the night “swallowing” the village. The AI version replaced that image with a more conventional expression describing darkness covering or surrounding the village.
Another example was “Her voice was a knife wrapped in silk.” Human translators retained the imagery of the knife and silk or reconstructed the contrast between softness and sharpness. The AI translation communicated the meaning as a soft voice with very sharp words, but removed the concrete images that created the literary effect.
The finding suggests that AI can understand the basic function of a metaphor without necessarily reproducing its aesthetic form. In other words, the difference is not always whether AI understands the metaphor, but how it chooses to represent that metaphor in the target language.
Narrative voice showed a similar pattern. Human translators maintained more of the narrator's emotional attitude, while AI tended to produce a more neutral and standardized version. The event remained understandable, but the relationship between narrator, character, and reader could become less expressive.
Cultural Nuances Can Become Generalized by AI
Cultural representation was another important difference.
When translating the expression “The parish priest greeted them after Sunday service,” human translators used more culturally specific alternatives. The AI version adopted a broader expression referring to religious leaders and Sunday worship.
This approach improves general readability, but it can also remove information about the particular religious institution represented in the original text.
The same issue appeared in forms of address such as “My Lord.” Human translators were more likely to preserve the social hierarchy and formality implied by the expression. AI used a more general form of address, maintaining the basic interaction while reducing some of the social information embedded in the original wording.
The research therefore suggests that cultural translation is not simply about finding an equivalent dictionary definition. Translators must also consider social relationships, cultural practices, historical context, and the function of an expression within the story.
AI Is Better Viewed as a Translation Partner
The findings do not suggest that AI has no role in literary translation. Instead, Gunawan Tambunsaribu of Universitas Kristen Indonesia argues for a complementary relationship between human expertise and AI technology.
AI can provide fluent and consistent translations, generate alternative wording, explore vocabulary, and accelerate the initial stages of translation. Human translators, however, remain particularly important when decisions involve ambiguity, metaphor, narrative voice, idiomatic meaning, and culturally specific expressions.
An ethical paraphrase of Tambunsaribu's central insight is that literary translation should not be judged only by fluency or basic semantic accuracy; preserving style and cultural representation is also essential to preserving the meaning of a literary work.
This perspective has practical implications for publishers, editors, translators, educators, and students. AI can be used to improve efficiency, but AI-generated literary translations should receive additional human review when they contain metaphors, idioms, narrative expressions, or culture-specific references.
Implications for Education and the Translation Industry
For universities and translation programs, the findings highlight the importance of teaching students to evaluate AI output critically rather than simply accepting fluent machine-generated text.
For publishers and professional translators, AI may serve as an efficient drafting and brainstorming tool. Human expertise remains essential for deciding whether a particular literary image should be retained, adapted, explained, or replaced with an equivalent expression in the target culture.
The study also cautions against treating its findings as a universal ranking of humans versus AI. Only three human translators and one AI model were examined, and the researchers deliberately selected complex literary passages containing figurative and cultural elements. The findings therefore describe qualitative patterns rather than statistically proven differences across all translators or AI systems.
Future research could compare more translators, literary works, language pairs, and AI models. Researchers could also test different prompting strategies to determine whether literary-aware or culture-aware instructions improve AI's ability to preserve stylistic and cultural elements.
Author Profile
Gunawan Tambunsaribu is affiliated with Universitas Kristen Indonesia and is the author of the study on human and artificial intelligence in literary translation. His research in this article focuses on literary translation, meaning, linguistic style, cultural representation, and human–AI interaction in translation. The published article does not specify an academic degree, so no degree is attributed to the author here.
0 Komentar