Research on the logical coherence of translations produced by AI technology - Taking the excerpt of the 2024 Government Work Report as an example
AI技术产出译文的逻辑连贯性研究——以《2024政府工作报告》片段为例

摘要

AI技术的不断更新迭代对翻译行业带来了巨大的质量和效率提升，同时也对译者提出了新挑战，通过机器翻译和AI工具可以完成语言上的字面转换，然而，通过AI产出的译文质量，尤其是逻辑连贯性方面，仍存在显著缺陷。逻辑性是语言至关重要的一个方面，中英两种语言之间存在着显著的逻辑层面的差异，因此在翻译中常常要做出必要的调整以确保文字的逻辑连贯。本文就当前国内市场上的语言模型“文心一言”和综合实力强大的ChatGPT产出的译文进行逻辑层面的分析。通过对比官方译文，发现AI产出的译文在逻辑上会出现重复、段落间和句间逻辑的理解谬误、词语搭配等偏差。这些发现一方面说明机器翻译仍然依赖于人工编辑与校对，另一方面也表明了机器翻译亟待改进。

Abstract

The continuous evolution of AI technology has significantly enhanced translation quality and efficiency within the industry, while simultaneously presenting new challenges for human translators. Although machine translation and AI tools demonstrate competence in literal linguistic conversion, their outputs exhibit substantial deficiencies in textual quality, particularly regarding logical coherence. As logical consistency constitutes a fundamental aspect of linguistic expression, the inherent disparities in logical frameworks between Chinese and English necessitate systematic adaptations during translation to ensure textual cohesion. This study conducts a critical analysis of logical inconsistencies in translations generated by two prominent language models in the domestic market, “Wenxin Yiyan” and the globally recognized ChatGPT. Through comparative analysis with authoritative reference translations, the research identifies recurrent issues in AI-generated outputs, including redundant expressions, misinterpretations of inter-paragraph and inter-sentential logical relationships, and collocational deviations. These findings not only reaffirm the indispensable role of human editorial intervention in machine translation workflows but also highlight critical directions for technological refinement in artificial intelligence translation systems.