Emotion Profiling in LLM-Based Literary Translation: Systematic Shifts Across MT and Post-Editing

2026-06-08

AI Summary

This study investigates whether large language models (LLMs) introduce systematic affective biases in literary translation and whether post-editing can realign their output with human translators’ affective norms. Leveraging a contemporary Italian science fiction corpus, the research compares raw LLM translations, post-edited versions, and professional human translations through fine-grained sentiment analysis combining lexicon-based and multilingual affective modeling approaches. The findings reveal, for the first time, that LLMs imprint distinct, model-specific affective “fingerprints” that systematically distort the author’s original emotional style. Although post-editing partially mitigates these deviations, it fails to fully restore affective fidelity. This work provides quantitative evidence and an evaluation framework for preserving authorial voice in machine translation, highlighting persistent challenges in aligning automated systems with human affective expression.

Abstract

This paper investigates whether LLM translations exhibit identifiable emotional profiles and how post-editing reshapes them toward human-like norms. We compare LLM translations of Margaret Atwood's Oryx and Crake with their post-edited versions and a human translation, using a large-scale corpus of contemporary Italian science-fiction as a baseline. We examine emotion through lexicon-based and multilingual modeling, conducting a fine-grained analysis of emotional variation across systems. We find that MT systems introduce model-specific and statistically significant emotional fingerprints across translations, leading to a limited preservation of an author's voice.

Problem

• emotion profiling
• literary translation
• large language models
• post-editing
• authorial voice

Innovation

• emotion profiling
• LLM-based translation
• post-editing
• authorial voice preservation
• multilingual emotion modeling