Can ChatGPT Break Through the ‘Bastion’ of Literary Translation? A Stylometric Study of Human vs. Machine Translation Variants

Date:

Research Design:

  • Comparative analysis of 7 human translations (HT) vs. 14 machine translations (MT)
  • Target texts: Two canonical literary essays
  • Methodology: Text similarity algorithms for stylometric measurement

Key Discoveries:

  • LLMs surpass NMT in stylistic adaptability
  • Human translators maintain 37% greater stylistic range than best-performing LLM
  • “Translation universals” manifest differently across HT/LLM/NMT

Implications:

  • Challenges in automating literary translation
  • Quantitative framework for assessing “translation style”
  • Benchmark for future LLM development in creative domains