본문 바로가기
  • Home

A Follow-up Study of Stylistic Differences between Human and Machine Translation with ChatGPT Added in the Mix

  • The Journal of Translation Studies
  • Abbr : JTS
  • 2023, 24(3), pp.539-561
  • DOI : 10.15749/jts.2023.24.3.017
  • Publisher : The Korean Association for Translation Studies
  • Research Area : Humanities > Interpretation and Translation Studies
  • Received : July 13, 2023
  • Accepted : September 19, 2023
  • Published : September 30, 2023

Chang-Soo Lee 1

1한국외국어대학교

Accredited

ABSTRACT

The present study explores whether new shifts have developed in the stylistic landscape of human vs. machine translation in the wake of ChatGPT’s arrival. For this purpose, it conducts a series of principal component analyses (PCAs) on a normalized frequency dataset comprising 67 morphological and syntactic linguistic features borrowed from Biber’s (1988) research on register variation. The dataset is derived from a corpus of Korean editorials from three Korean newspapers, their human English translations, and English translations generated by four machine translation systems (Papago, Google, DeepL, ChatGPT), including ChatGPT’s self-proofread versions. The analyses indicate that human and machine translation remain distinctly differentiated in terms of style, as demonstrated in previous studies. However, among the machine translation systems, ChatGPT, both in its translations and self-proofread versions, deviates significantly from the others. A closer examination of the linguistic features strongly associated with ChatGPT reveals that this difference can be attributed to the model’s intrinsic preference for a formal, written style. Notably, there are no substantial stylistic divergences between ChatGPT’s translations and its self-proofread versions.

Citation status

* References for papers published after 2023 are currently being built.