前两周，ChatGPT大火，即便对AI不感兴趣的我，看着一个个ChatGPT新闻成为头条，还是不免感到焦虑，感觉似乎不做点什么就会成为时代的弃儿。
我曾看到不少推文说ChatGPT擅长进行科研论文的翻译和润色，并夸它翻译得如何如何地道准确。我在过去的写作经历里使用过不少翻译软件，也相当好奇ChatGPT在语言翻译上究竟能有几把刷子。
ChatGPT在翻译时需有提示词（Prompt）作为引导才能引导系统进行翻译。所以，提示词本身也会影响翻译输出的质量。为此，我想应当寻找一个适合提问的语法。在网上搜索一通后，我突然醒悟过来。既然ChatGPT是人工智能，那我能不能直接问ChatGPT该如何提问它呢？(似乎有点绕口)
Round1
医学术语翻译PK
Round2
-原文摘要片段-
这是中文摘要：
方法：基于中国健康与养老调查2011-2018(CHARLS2011-2018)数据，共纳入12038例样本，将认知水平划分为4个等级,对不同认知人群生存率进行LogRank检验,采用多因素Cox回归模型调整混杂因素，探讨不同认知人群的死亡风险差异。
这是中文摘要自带的英文：
Methods：BasedonCHARLS2011-2018,atotalof12038sampleswereincludedandthecognitivelevelwasdividedintofourlevels.LogRanktestwasperformedonthesurvivalratesofdifferentcognitivegroups.ThemultivariateCOXregressionmodelwasusedtoadjustthemixedfactorstoexplorethedifferencesinthemortalityriskofdifferentcognitivegroups.
1
ChatGPT译文
Methods:BasedondatafromtheChinaHealthandRetirementLongitudinalStudy(CHARLS)2011-2018,atotalof12,038participantswereincluded,andcognitivelevelsweredividedintofourcategories.ThesurvivalratesofdifferentcognitivegroupswerecomparedusingtheLogRanktest,andamultivariateCoxregressionmodelwasusedtoadjustforconfoundingfactorsandinvestigatedifferencesinmortalityriskbetweendifferentcognitivegroups.
2
DeepL译文
METHODS:BasedondatafromtheChinaHealthandAgingSurvey2011-2018(CHARLS2011-2018),atotalof12038sampleswereincluded,cognitivelevelwasclassifiedinto4classes,andLogRanktestswereperformedtoexaminethesurvivalratesofdifferentcognitivepopulations,andmulti-factorCoxregressionmodelswereusedtoadjustforconfoundingfactorsandexplorethedifferencesinmortalityriskamongdifferentcognitivepopulations.
简要评述
DeeplvsChatGPT
ChatGPT回答：
Methods:DatafromtheChinaHealthandRetirementLongitudinalStudy(CHARLS)spanningfrom2011to2018wereanalyzed,with12,038participantsincluded.Thecognitivelevelofparticipantswasclassifiedintofourcategories.TheLog-Ranktestwasusedtocomparesurvivalratesamongdifferentcognitivegroups,andaCoxproportionalhazardmodelwasusedtoadjustforpotentialconfoundingfactorsandexploredifferencesinmortalityriskbetweenthecognitivegroups.
我还是不过瘾，我又问能不能按照医学第一刊物NEJM的格式给我写出摘要：PleasepolishtheabstractaccordingtothestyleoftheNEJMjournal？
ChatGPT的回答：
Methods:WeanalyzeddatafromtheChinaHealthandRetirementLongitudinalStudy(CHARLS)spanningfrom2011to2018.Weincluded12,038participantsandclassifiedtheircognitivelevelsintofourcategories.WetestedthesurvivalratesofthedifferentcognitivegroupsusingtheLog-RanktestandusedtheCoxproportionalhazardmodeltoadjustforconfoundingfactorsandinvestigatedifferencesinmortalityriskbetweenthecognitivegroups.
真心不错！ChatGPT可以提供多个版本，总有一个版本可能是你想要的！第三人称不要，可以换第一人称视角来介绍。这恐怕是一般的软件和人工译稿没法直接达成吧！
（图片截自B站up主ZhanI的视频）
参考资料：
[1]TranslatorAPreliminaryStudy.WenxiangJiao,WenxuanWang,Jen-tseHuanget.al.Arxiv.2301.08745