Literary voice, cultural meaning, and human revision in large language model translation from 2022 to 2026

Large Language Models (LLMs) produce literary translations that read fluently, yet fluency reveals little about whether a narrator's voice, a poem's indeterminacy, or a culturally loaded reference survives. This systematic review examined how literary quality has been defined and measured in empirical studies of LLM literary translation, how far different kinds of evidence converge, and under which conditions prompting and human revision alter the preservation of literary features in translated texts. Semantic Scholar and Crossref searches, journal-targeted queries, and citation chasing covered January 2022 to September 5, 2026, and eligibility was restricted to journal articles from Scopus-listed sources. Forty-one studies were included, 28 extracted from full texts and 13 from abstracts, and each was appraised on nine validity domains without a summed score. The evidence is concentrated on Chinese and Arabic source texts, English targets (28 studies), and GPT models (37 studies), and 28 studies relied on a single evaluation method. Accuracy and fluency often matched or exceeded published human translations, whereas narrator differentiation, culturally framed stances, symbolic references, and taboo forces were compressed or neutralized in better-controlled designs. Overlap metrics broadly agreed with expert or reader judgments in only one of the seven studies reporting both, and reader backgrounds changed rankings. Prompt effects were model-specific and sometimes traded stylistic gains for unsupported additions, while human revision appeared only in case- and practice-based designs. A five-layer evaluation framework is proposed in which claims about voice and culture require evidence at the layers they invoke, together with research priorities for Indonesian and other underrepresented literature.

Authors

Publication Details

Journal
Journal of Language and Literature Inquiry
Published
2026-11-15
DOI
https://doi.org/10.55942/jlli.v1i2.2364
Primary Topic
Translation Studies and Practices
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Literary voice, cultural meaning, and human revision in large language model translation from 2022 to 2026

Devina Antania Putri Maharianto
Journal of Language and Literature Inquiry
Translation Studies and Practices
article

Literary voice, cultural meaning, and human revision in large language model translation from 2022 to 2026

Devina Antania Putri Maharianto
article en

Abstract

Large Language Models (LLMs) produce literary translations that read fluently, yet fluency reveals little about whether a narrator's voice, a poem's indeterminacy, or a culturally loaded reference survives. This systematic review examined how literary quality has been defined and measured in empirical studies of LLM literary translation, how far different kinds of evidence converge, and under which conditions prompting and human revision alter the preservation of literary features in translated texts. Semantic Scholar and Crossref searches, journal-targeted queries, and citation chasing covered January 2022 to September 5, 2026, and eligibility was restricted to journal articles from Scopus-listed sources. Forty-one studies were included, 28 extracted from full texts and 13 from abstracts, and each was appraised on nine validity domains without a summed score. The evidence is concentrated on Chinese and Arabic source texts, English targets (28 studies), and GPT models (37 studies), and 28 studies relied on a single evaluation method. Accuracy and fluency often matched or exceeded published human translations, whereas narrator differentiation, culturally framed stances, symbolic references, and taboo forces were compressed or neutralized in better-controlled designs. Overlap metrics broadly agreed with expert or reader judgments in only one of the seven studies reporting both, and reader backgrounds changed rankings. Prompt effects were model-specific and sometimes traded stylistic gains for unsupported additions, while human revision appeared only in case- and practice-based designs. A five-layer evaluation framework is proposed in which claims about voice and culture require evidence at the layers they invoke, together with research priorities for Indonesian and other underrepresented literature.

Journal of Language and Literature InquiryVol. 1(2)
Translation Studies and Practices
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon