GPT-4 vs. Human Translators: A Comprehensive Evaluation of Translation Quality Across Languages, Domains, and Expertise Levels

TL;DR

This study compares GPT-4's translation quality with human translators across languages and domains, revealing GPT-4's strengths and weaknesses, and providing insights into LLMs' current capabilities and limitations in translation tasks.

Contribution

First comprehensive evaluation of GPT-4 against human translators across multiple languages and domains, analyzing systematic differences and translation quality.

Findings

GPT-4 performs comparably to junior translators in total errors.

GPT-4's performance declines in resource-poor languages and domains.

GPT-4 tends to produce literal translations, while humans sometimes overthink background context.

Abstract

This study comprehensively evaluates the translation quality of Large Language Models (LLMs), specifically GPT-4, against human translators of varying expertise levels across multiple language pairs and domains. Through carefully designed annotation rounds, we find that GPT-4 performs comparably to junior translators in terms of total errors made but lags behind medium and senior translators. We also observe the imbalanced performance across different languages and domains, with GPT-4's translation capability gradually weakening from resource-rich to resource-poor directions. In addition, we qualitatively study the translation given by GPT-4 and human translators, and find that GPT-4 translator suffers from literal translations, but human translators sometimes overthink the background information. To our knowledge, this study is the first to evaluate LLMs against human translators and analyze the systematic differences between their outputs, providing valuable insights into the current state of LLM-based translation and its potential limitations.