BLEU Score Dataset of Chinese-English Financial News Translation by 3 Domestic Large Language Models (Reasoning & Non-Reasoning Dual Modes, Context Isolation Environment)

This dataset is the supporting experimental measured data of the undergraduate thesis A BLEU-Based Quantitative Comparison of AI Translation Tools for Chinese-English Financial News Texts, with the core being the quantitative evaluation results of Chinese-English financial news translation quality by large language models (LLMs).In this study, 50 authoritative Chinese-English parallel financial news sentences published by China Daily in 2025 were taken as translation samples, and three mainstream domestic LLMs in China, namely Doubao, DeepSeek and Yuanbao, were selected as the research objects.

In a strictly context-isolated experimental environment (dialogue memory function disabled, independent dialogue for each sample to eliminate contextual interference), the English translations of the models under Reasoning Mode and Non-Reasoning Mode were obtained respectively. With the internationally accepted BLEU (Bilingual Evaluation Understudy) as the core indicator, the translation quality score corresponding to each translation was calculated.This dataset includes the full original BLEU score data of 50 samples from 6 experimental groups (3 models × 2 generation modes), as well as the corresponding descriptive statistical results including mean, standard deviation, 95% confidence interval, variance and coefficient of variation.