机器有机器的理解:机器翻译中理解问题的哲学思考  

Understand Machine's Understanding:A Philosophy Thinking on Understanding in Machine Translation

作　　者：沈宏梁 王新 罗晖 SHEN Hongliang;WANG Xin;LUO Hui(Princeton Institute of Intelligence Research and Application,Princeton,NJ 08540,USA)

机构地区：[1]普林斯顿智能探索与应用研究所,美国普林斯顿NJ 08540

出　　处：《自然辩证法通讯》2025年第3期35-41,共7页Journal of Dialectics of Nature

摘　　要：本文讨论了关于机器翻译是否需要人类那样的理解这个哲学问题。尽管在科学和哲学上对什么是人的理解还没有解决,本文提出了现代最先进的机器翻译方法的成功其实是基于人类语言的可组合性。只要机器可以从双语训练语料库“学习”和“理解”两种语言的可组合性以及它们之间的对应关系,就可以实现高质量的机器翻译。换句话说,机器可以有不同于人类的对于人类语言的理解方式。This paper discusses the philosophical question on whether understanding in human style is necessary for machine translation.Although it remains unsolved what human understanding exactly means in science and in philosophy,this paper points out the success of state of the art machine translation methods roots in compositionality of human languages.As long as the machine can“learn”and“understand”the mapping and the compositionalities of two languages from bilingual training corpora,machine translation can achieve good quality.In other words,machines can have their own understanding of human languages,different from human understanding.

关 键 词：机器翻译 理解 歧义 

分 类 号：H085[语言文字—语言学] N031[自然科学总论—科学技术哲学]