Abstract:
The lecture explores how translation workflows affect performance in both human and AI literary translation.  Large language models (LLMs) perform well in general translation but often struggle with literary nuance and coherence. Professional human translations are compared with two multi-agent LLM systems: one modeled on human practice with drafting and revision phases, and another redesigned for LLM capabilities with specialized agents for planning, style, and coherence. Expert evaluations show that both AI systems achieved accuracy comparable to human translators. The LLM-oriented workflow produced more stylistically rich and fluent translations, occasionally with creative additions, whereas the human-modeled workflow was concise but less cohesive. The findings suggest that workflows optimized for LLM capabilities can yield exceptional literary translations and, in certain aspects, surpass human performance.

Biography:

Kanglong Liu is Associate Professor in the Department of Language Science and Technology at The Hong Kong Polytechnic University. His research focuses on empirical approaches to translation studies, translation pedagogy, corpus-based translation research, and Hongloumeng translation studies. His work has appeared in leading journals such as Target, Perspectives, Lingua, Language Sciences, International Journal of Specialised Translation, System, International Review of Applied Linguistics in Language Teaching, and Digital Scholarship in the Humanities. He serves as Associate Editor of Translation Quarterly, Humanities and Social Sciences Communications (SSCI and AHCI), and Heliyon (SCI). He is the author of Corpus-Assisted Translation Teaching: Challenges and Issues (Springer, 2020) and co-editor of several volumes, including Translation and Interpreting in the Age of COVID-19 (Springer, 2023), Dream of the Red Chamber: Literary and Translation Perspectives (Routledge, 2023), Corpora in Interpreting Studies: East Asian Perspectives (Routledge, 2023), and Translation Studies in the Age of Artificial Intelligence (Routledge, 2025).