Garuda - Garba Rujukan Digital

POPULAR OR POPULER? COMPARING AI AND HUMAN TRANSLATION OF NONCE WORDS IN WICKED’S INDONESIAN SUBTITLES Abdillah, Taufik Eryadi; Fanny Puji Rakhmi; Ina Sukaesih; Farizka Humolungo; Septina Indrayani
CALL Vol. 7 No. 2 (2025): CALL
Publisher : Universitas Islam Negeri Sunan Gunung Djati

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.15575/call.v7i2.51205

This study investigates the process of translating nonce words in Indonesian subtitles of the musical fantasy film Wicked, focusing on the challenges posed by highly creative and humorous expressions that remain underexplored in audiovisual translation studies. Set in the magical land of Oz, the film incorporates playful and imaginative expressions like Galindafied and braverism, which present unique translation difficulties. The study employs content analysis to compare human-generated subtitles from Apple TV with AI-generated subtitles produced by ChatGPT. While both human and AI translations tend to convey the general meaning of the nonce words, they fail to capture the stylistic and humorous nuances present in the source language. Human translations can be literal or omit creative expressions entirely, suggesting that neither method fully encapsulates the inventiveness and playfulness of the source language. This study underscores the importance of developing more adaptable strategies for translating highly creative audiovisual texts.

Investigating Romantic-Tone Transfers in the Human Translation Compared to AI Subtitling: A Case Study of the Dilan 1990 Film Rakhmi, Fanny Puji; Abdillah, Taufik Eryadi; Humolungo, Farizka; Sukaesih, Ina; Indrayani, Septina
Journal of Language and Literature Studies Vol. 6 No. 1 (2026): March
Publisher : Lembaga Penelitian dan Pemberdayaan Masyarakat (LITPAM)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36312/jolls.v6i1.3570

This paper examines how romantic tone is carried across languages in film subtitling, where text must be brief, well-timed, and easy to read. We focus on Dilan 1990, a popular Indonesian teen romance known for playful, indirect lines that rely on inference and rhythm. Today, many studios generate first-pass subtitles with neural machine translation and then rely on human post-editors to refine style and timing. We ask how well AI handles this kind of subtle, affect-rich dialogue and where human editing still adds value. We compare 24 well-known lines addressed by Dilan to Milea in two versions: the official English subtitles and outputs produced by ChatGPT under the same line-level constraints. Using a practical set of translation techniques (e.g., modulation, compression, adaptation) and a tone rubric (playfulness/coyness, warmth, persona/rhythm), we perform a line-by-line analysis. The human subtitles tend to keep implicature and pace through concise, idiomatic choices that fit character and reading speed. The AI versions are fluent but more likely to explain subtext or lengthen the line, which can blunt teasing and shift the scene’s mood. AI can match human choices when the source line is already compact and direct. Where meaning depends on ellipsis, metaphor, and micro-timing, human post-editing remains crucial. We close with practical guidelines for NMT-plus-post-editing workflows in romance subtitling.

ANALYSIS OF FLOUTING MAXIMS AND CONVERSATIONAL IMPLICATURE TRANSLATION STRATEGIES IN ONE PIECE (LIVE-ACTION SERIES 2023) SUBTITLE Roselin Ahmadi; Farizka Humolungo; Fanny Puji Rakhmi
LinguAmerta Vol. 1 No. 2 (2024): Vol. 1 No. 2 2024
Publisher : Politeknik Negeri Jakarta

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32722/jla.v2i1.7251

This study examines the flouting of Cooperative Principle maxims in conversational implicature and Gottlieb’s translation strategies used in the subtitles of One Piece (Live-Action Series 2023). This study aims to identify which conversational maxims are flouted, Gottlieb’s translation strategies employed, and how these implicatures are translated into the target language. This study used a qualitative descriptive method with document analysis as the data collection technique. The results showed that all four conversational maxims were flouted, and some implicatures flouted multiple maxims simultaneously. Most implicatures were translated into the same implicature in the target text, while a small number were translated explicitly and have undergone explicitation. The most frequently used translation strategy was Paraphrase, followed by Condensation, Transfer, Dislocation, Decimation, Expansion, and Deletion. Most of the implicatures translated explicitly use the Paraphrase strategy to ensure that the utterances are clear and align with the conversation.

A comparative evaluation of ChatGPT and human translation of Sundanese cultural terms Ina Sukaesih; Fanny Puji Rakhmi; Farizka Humolungo; Septina Indrayani; Taufik Eryadi Abdillah
Studies in English Language and Education Vol. 13 No. 2 (2026)
Publisher : Universitas Syiah Kuala

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24815/siele.v13i2.1108

This study addresses a gap in low-resource language translation research by comparing how ChatGPT and a human translator render Sundanese cultural terms into English and evaluating both outputs using established translation quality dimensions. A qualitative descriptive-comparative design was used. A total of 109 cultural terms were extracted from the Sundanese short story Jampe Ngimpi, together with their two English versions (human and ChatGPT). A panel of bilingual Sundanese–English raters familiar with Sundanese culture independently assessed each item for accuracy, acceptability, and readability using Nababan’s Translation Quality Assessment rubric (three-point scale), followed by focus group discussions to reach consensus. The data are analyzed by combining numerical averages with brief qualitative explanations of why certain parts are strong or weak based on the three parameters. The findings show differences in performance across translation quality dimensions. ChatGPT records relatively strong results in accuracy (2.5) and acceptability (2.5), while human translators show a slight advantage in readability (2.6), particularly for terms requiring deeper cultural contextualization. ChatGPT performs better with culturally straightforward items that have clear English equivalents but tends to produce literal calques or transfer Sundanese forms with minimal explanation when faced with culturally specific terms, which can reduce clarity and obscure pragmatic meaning. In contrast, human translators more often apply brief glosses, paraphrasing, and descriptive equivalents that preserve cultural meaning while improving reader comprehension. The results indicate that ChatGPT is effective for initial drafting, while human revision remains essential for ensuring cultural accuracy and contextual appropriateness in Sundanese-English translation.

Language Learning Strategies Among Vocational Generation Z Students: Insights from A Digital Native Perspective Nidia Sofa; Fanny Puji Rakhmi; Iis Mariam; Mawarta Onida
Journal of English Language Studies Vol 10, No 2 (2025): Available Online in September 2025
Publisher : English Department - University of Sultan Ageng Tirtayasa

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.62870/jels.v10i2.32504

This study explores the English language learning strategies employed by vocational Generation Z students at Politeknik Negeri Jakarta, focusing on their integration of digital-native characteristics within an English for Specific Purposes (ESP) context. Employing a quantitative survey design, data were collected from second-semester vocational students through two adapted instruments: Oxford’s Strategy Inventory for Language Learning (SILL) and Teo’s Digital Natives Assessment Scale (DNAS). A total of 108 students participated in the study. Data analysis revealed that students employed direct, indirect, and digital-native strategies at comparably high frequencies, with no statistically significant differences among the strategy types (p > 0.05). At the subscale level, instant gratification or rewards was most prominent, followed by metacognitive and technology use, whereas graphic communication was least used, with compensation and affective relatively lower. It is recommended that ESP educators should blend strategies within each task: begin with brief targeted language practice (direct), include a short plan/monitor step or peer check (indirect/metacognitive), and complete the task with approved digital support in order to optimize ESP learning for vocational education contexts. Future work should validate the adapted digital-native scale in local contexts, replicate across diverse ESP settings, and link reported strategies to objective learning outcomes using longitudinal and mixed-methods designs.

Investigating Romantic-Tone Transfers in the Human Translation Compared to AI Subtitling: A Case Study of the Dilan 1990 Film Rakhmi, Fanny Puji; Abdillah, Taufik Eryadi; Humolungo, Farizka; Sukaesih, Ina; Indrayani, Septina
Journal of Language and Literature Studies Vol. 6 No. 1 (2026): March
Publisher : Lembaga Penelitian dan Pemberdayaan Masyarakat (LITPAM)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36312/jolls.v6i1.3570

This paper examines how romantic tone is carried across languages in film subtitling, where text must be brief, well-timed, and easy to read. We focus on Dilan 1990, a popular Indonesian teen romance known for playful, indirect lines that rely on inference and rhythm. Today, many studios generate first-pass subtitles with neural machine translation and then rely on human post-editors to refine style and timing. We ask how well AI handles this kind of subtle, affect-rich dialogue and where human editing still adds value. We compare 24 well-known lines addressed by Dilan to Milea in two versions: the official English subtitles and outputs produced by ChatGPT under the same line-level constraints. Using a practical set of translation techniques (e.g., modulation, compression, adaptation) and a tone rubric (playfulness/coyness, warmth, persona/rhythm), we perform a line-by-line analysis. The human subtitles tend to keep implicature and pace through concise, idiomatic choices that fit character and reading speed. The AI versions are fluent but more likely to explain subtext or lengthen the line, which can blunt teasing and shift the scene’s mood. AI can match human choices when the source line is already compact and direct. Where meaning depends on ellipsis, metaphor, and micro-timing, human post-editing remains crucial. We close with practical guidelines for NMT-plus-post-editing workflows in romance subtitling.

Penerapan Teknik Penerjemahan dalam Peningkatan Kompetensi Penerjemah PT Indo Lingua Translocalize Ina Sukaesih; Farizka Humolungo; Septina Indrayani; Fanny Puji Rakhmi; Taufik Eryadi Abdillah
Mitra Akademia: Jurnal Pengabdian Masyarakat Vol 9 No 1 (2026): Mitra Akademia: Jurnal Pengabdian Masyarakat
Publisher : Pusat Penelitian dan Pengabdian kepada Masyarakat (P3M) Politeknik Negeri Jakarta

Show Abstract | Download Original | Original Source | Check in Google Scholar

This community service program aimed to implement research findings on Molina and Albir’s (2002) translation techniques among translators at PT Indo Lingua Translocalize. The partner is a language service company established in 2009, specializing in Indonesian translation, offering services such as translation, interpreting, and complementary services, including subtitling, desktop publishing (DTP), voice-over, copy-editing, reviewing, linguistic validation, and CAT tools (SDL Trados) distribution. The program was initiated in response to the partner’s need to enhance the quality of literary translation, particularly in terms of accuracy, acceptability, and readability of cultural elements. The program was conducted through theoretical input, interactive discussions, translation practice, and evaluation using pre-tests and post-tests. The method employed was Participatory Action Research (PAR) with a participatory empowerment approach. The findings indicate improved participants’ understanding of translation techniques and their ability to appropriately apply these techniques to literary texts. Participants also reported positive responses regarding the relevance of the materials, clarity of delivery, and professional benefits. This program contributes to strengthening translators’ competence and supports the preservation of local languages through research-based translation practices.

READABILITY ASSESSMENT OF AI-DRIVEN INDONESIAN JOURNALISTIC TRANSLATIONS: COMPARING CHATGPT AND DEEPL PERFORMANCE Farizka Humolungo; Fanny Puji Rakhmi; Ina Sukaesih; Ince Dian Aprilyani Azir; Talenta Gloria Napitupulu
Jurnal Kata Vol. 10 No. 1 (2026): Jurnal Kata : Penelitian tentang Ilmu Bahasa dan Sastra
Publisher : LLDIKTI Wilayah X

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.22216/7ngabs09

This study investigates the readability of Indonesian to English translations of journalistic texts produced by ChatGPT and DeepL across the domains of law, politics, and environment using articles from Tempo and Kompas. A qualitative descriptive comparative design was employed. Six source texts were translated under controlled procedures, with standardized prompts for ChatGPT and standard input for DeepL. Two expert raters evaluated segment level readability using a three point rubric adapted from prior work. Before scoring, the raters reviewed and aligned category descriptors to support shared interpretation. Readability scores were summarized by outlet and topic, and rater comments from a focused discussion were used to contextualize patterns. The findings show small and patterned differences between systems. One rater tended to rate ChatGPT as more readable, while the other found broad parity. Readability was highest for environmental reporting and most challenging for legal texts that contain dense institutional references. Kompas items were generally read as more accessible than Tempo items. The results point to domain sensitive strengths for each system and suggest targeted post editing as a practical lever to support clarity for public facing news.

Title

Found 8 Documents
Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title Search

Found 8 Documents Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title

Found 8 Documents
Search