Azzahra, Adinda Ayu (2026) A STUDY OF HUMAN AND AI PERFORMANCES IN CATEGORIZING AND TRANSLATING SOCIAL DEIXIS IN PRAMOEDYA ANANTA TOER’S BUMI MANUSIA. Masters thesis, Fakultas Ilmu Budaya.

	Text ML142 ADINDA AYU AZZAHRA 13020324410002 OK_1.pdf Download (2MB)
	Text ML142 ADINDA AYU AZZAHRA 13020324410002 OK_2.pdf Download (1MB)
	Text ML142 ADINDA AYU AZZAHRA 13020324410002 OK_3.pdf Download (1MB)
	Text ML142 ADINDA AYU AZZAHRA 13020324410002 OK_4.pdf Download (1MB)
	Text ML142 ADINDA AYU AZZAHRA 13020324410002 OK.pdf Restricted to Repository staff only Download (2MB)

Abstract

Today, translation, which used to be conducted by human translators only, is a task
which can be automated, not only by automatic machine translation system, but also
by AI tools. This study aimed to: 1) Examine the accuracy of AI agents in
identifying and categorizing social deixis, and 2) Examine the accuracy of AI agents
in translating social deixis based on their selection of Molina & Albir's (2002)
Translation Techniques in Pramoedya Ananta Toer's Bumi Manusia. The data
consist of translations produced by human translator in its English version as the
reference, also results generated by ChatGPT and Gemini collected using
documentation technique and analyzed using qualitative-quantitative mixed
methods. This study identified the categorization of social deixis based on
Levinson’s (1983) theory. To ensure consistency in the identification process, interrater agreement was applied and reached high level of reliability (97%). After
establishing the data, AI agents were used to categorize the types of social deixis.
The same expressions were then translated by the AI for comparison. The results
show that the human translator and ChatGPT mostly categorize social deixis as
absolute, while Gemini more often categorizes it as relational. The study then
analyzed the translation techniques framework proposed by Molina & Albir (2002)
that are selected by each agent in rendering social deixis into the target language.
The results show that there are 15 techniques are identified and the AI agents have
a low agreement with human translator. Through this process, this research provides
a coherent comparison of the patterns used by human and AI translators in handling
socially marked language. In addition, the agreement is higher in the categorization
of social deixis than in its translation, although each AI agent demonstrates
comparable patterns in the application of translation techniques.
Keywords: social deixis, artificial intelligence, translation, bumi manusia

Actions (login required)

View Item