chatGPT是一个基于GPT-3模型的聊天机器人。GPT-3是由OpenAI开发的一种大型语言模型，能够生成人类般自然的语言文本。chatGPT使用这一模型来回答用户的问题，并且能够进行复杂的对话，在2022年11月30日正式公开。 chatGPT还具有情感分析能力，能够识别用户的情绪并做出相应的回应。它可以用于各种应用场景，包括客服机器人、在线问答系统、社交媒体聊天机器人等。
因为其有出色的（目前来看强于谷歌翻译）翻译性能，因此我尝试将原先的中文版论文翻译为英文，并将谷歌翻译版本和GPT版本进行比较。
ROUND1
原文：近年来，基于RNA介导的核酸内切酶CRISPRCas系统能在基因组特定位点引入DNA或RNA的链缺口，诱发宿主启动自身的防御机制修补缺口，大幅提高了同源重组效率，减少对筛选标记的依赖，实现多个物种的多基因编辑
Google翻译版本：In recent years, the RNA-mediated endonuclease CRISPRCas system can introduce a DNA or RNA strand gap at a specific site in the genome, inducing the host to initiate its own defense mechanism to repair the gap, greatly improving the efficiency of homologous recombination and reducing the need for screening markers. Dependence on multiple gene editing in multiple species.
GPT翻译版本：In recent years, CRISPRCas systems, which are RNA-mediated nucleic acid endonucleases, have been able to introduce DNA or RNA chain gaps at specific genomic loci, inducing the host's own defense mechanisms to repair the gaps, greatly improving the efficiency of homologous recombination and reducing reliance on screening markers, enabling multi-gene editing in multiple species.
可以看到GPT翻译的版本将句子的重点CRISPRCas系统提前，定于后置，这样读起来更加通顺，并且结尾部分也能够正确翻译
ROUND2
原文：StringTie继承于Cufflinks, 在准确性方面有了较大提升, 且可以通过输入数据库比对注释结果提高在已知基因区域的准确性, 在组装的过程中会计算每个基因及可变剪切的表达水平.
Google翻译版本：StringTie is inherited from Cufflinks, which has greatly improved its accuracy, and can improve the accuracy of known gene regions by inputting database comparison annotation results. During the assembly process Expression levels for each gene and alternative splicing are calculated.
GPT翻译版本：StringTie is an improved version of Cufflinks in terms of accuracy. It can improve the accuracy of known gene regions by inputting database comparison results. During the assembly process, it calculates the expression levels of each gene and alternative splicing.
这个句子里GPT翻译的结果似乎对原文进行了理解和加工，把原文的前两节合并为一句更加简练，并且句意不变，但是第二句明显少了“注释”这个单词。可以看出，GPT相比谷歌的翻译有语义分析的步骤，从单词与原文并不完全是一一对应的。
ROUND3
原文：由于组装算法的局限或变异位点分布不均匀，单纯使用全基因组测序组装的单体型都是局部的、片段化的。借助遗传信息分离同源区域的基因组数据， 再将每个区域组装成单体型，是目前解决高杂合物种组装最成功的方法。
Google翻译版本：Due to the limitations of the assembly algorithm or the uneven distribution of mutation sites, the haplotypes assembled solely by whole genome sequencing are localized and fragmented. Using genetic information to separate genomic data of homologous regions, and then assembling each region into a haplotype, is currently the most successful method to solve the assembly of highly heterozygous species.
GPT翻译版本：Due to the limitations of assembly algorithms or uneven distribution of polymorphic sites, the singleton type assembled solely by whole genome sequencing is local and fragmented. Using genetic information to separate homologous regions of genomic data and then assembling each region into a singleton type is currently the most successful method for solving high heterozygous species assembly.
这两个版本的翻译结果基本一致，除了GPT翻译的“单体型”翻译错误。
总结：目前chatGPT还不适合用来进行科研论文写作，一是在一些专业名词上容易出现翻译不准确，二是在翻译时可能会丢失关键的字词。当然当论文中有拿不准的一些句式或是想要探索更好的表达方式时，该工具可以提供一些思路。
转载本文请联系原作者获取授权，同时请注明本文来自张帅科学网博客。
链接地址：https://wap.sciencenet.cn/blog-3425594-1366824.html?mobile=1
收藏
