BLOG · 22/01/2023

我们为什么还是要用人脑学英语

使用 3 个主流翻译平台与 1 个语言大模型的翻译功能,简单用一段话检验机器翻译的流畅性与准确性。

从我最近在读的书中顺藤摸瓜扒拉到的文献摘要为例:

This study aimed to replicate and extend Kerckhoff and Davis’ (1962) findings concerning the effects of value consensus and need complementarity upon mate selection. Samples of “steadily attached” couples from two state universities reported on their courting relationships in both the fall (time1) and the spring (time2) of the school year. Results for 330 couples are presented. The previous findings, that value consensus and need complementarity would enhance progress toward permanence, in the present study received only marginal support. Moreover, the length of the couples’ relationship did not interact with these variables in accordance with the Kerckhoff-Davis study. Additional analyses of other Time time1 paper-and-pencil compatibility indices also failed to relate substantially to the Time time2 changes. The best predictors were the partners’ own Time time1 feelings about their relationship. Various implications of these results are discussed.

Progress Toward Permanence in Courtship: A Test of the Kerckhoff-Davis Hypotheses

Google翻译

一号选手是来自大厂的在线翻译:

流畅性不佳,逐字翻译而不得要领:

The previous findings, that value consensus and need complementarity would enhance progress toward permanence, in the present study received only marginal support.
先前的研究结果,即重视共识和需要互补性将促进持久性的进步,在本研究中仅获得边际支持

准确度不佳:

Moreover, the length of the couples’ relationship did not interact with these variables in accordance with the Kerckhoff-Davis study.
此外,根据 Kerckhoff-Davis 的研究,夫妻关系的长短与这些变量没有相互作用。

原文的意思是夫妻关系的长度与Kerckhoff-Davis的研究中的变量无关,翻译——完全错误!

DeepL

二号选手是来自德国的,号称第一的翻译服务:

流畅性高于谷歌,没有把marginal support直接翻译为边际支持,而是理解了意思后翻译为有限的支持

但准确性依然不佳:

Moreover, the length of the couples’ relationship did not interact with these variables in accordance with the Kerckhoff-Davis study.
此外,根据Kerckhoff-Davis的研究,夫妻关系的长度与这些变量没有相互影响。

这句话依然没有翻译对。

有道翻译

三号选手是来自国内的有名翻译服务:

流畅性不佳,和谷歌一样逐字翻译:

The previous findings, that value consensus and need complementarity would enhance progress toward permanence, in the present study received only marginal support.
在本研究中,先前的发现,即重视共识和需要互补性将促进迈向持久性的进展,仅获得边际支持

此外,准确度不佳:

Moreover, the length of the couples’ relationship did not interact with these variables in accordance with the Kerckhoff-Davis study.
此外,根据科克霍夫-戴维斯的研究,夫妻关系的长短与这些变量没有相互作用。

有道翻译似乎倾向于把人名都音译为汉字,个人很不喜欢,因为这样的翻译没有什么信息的转移,却很影响我溯源研究。

ChatGPT

最后一位选手是来自 OpenAI 的语言大模型:

流畅性仍然欠佳(虽然比谷歌有道翻译的边际支持好理解一些):

The previous findings, that value consensus and need complementarity would enhance progress toward permanence, in the present study received only marginal support.
先前的发现,即价值共识和需求互补将增强永久性的进展,在本研究中只得到了边缘支持

准确度优于传统的翻译服务:

Moreover, the length of the couples’ relationship did not interact with these variables in accordance with the Kerckhoff-Davis study.
此外,夫妇关系的长度与这些变量并没有根据 Kerckhoff-Davis 研究进行互动。

意思上终于正确了,尽管把Interact with直接翻译为进行互动很拗口。

更新Prompt,要求流畅而准确地翻译

遗憾的是,流畅性并没有变好:

The previous findings, that value consensus and need complementarity would enhance progress toward permanence, in the present study received only marginal support.
先前的研究发现,价值共识和需求互补将增强永久性的进展在本研究中只得到了边缘性的支持

综上,简单列一个表格:

翻译服务流畅性准确性
Googlexx
DeepLx
有道xx
ChatGPTx

综上所述,经过四个翻译服务的简单调查,目前没有任何一个在线服务可以同时兼顾流畅性与高度的准确性进行翻译。

再退一步,目前地球上没有任何服务可以把你复杂的阅读源:网页、PDF、照片、表格…直接快速到无感地翻译为流畅而可信的中文。

翻译服务的确好用,你能用 2%的努力获取 80%的可信信息。但是对于追求正确性与真实性的使用需求来说,如学术阅读与新闻检验而言,最后 20%至关重要,值得付出 98%的时间精力。

用人脑学习英语,在今日还是有必要的。

旅途碎碎念 part

在旅途过程中读完了一本书,准备挑一些有趣的东西写一篇 blog ,结果考据时意外发现翻译服务的不靠谱,再加上天天都在玩时间不太够,所以简单写一篇 ww

这几天向南走,去了阳光充沛的蒙德(语言+地理梗)。

不会下象棋的我抱着薯片看了好久街头热闹的象棋局,去了海边与湿地捡到一根一米多长的硕大芦苇边走边乱挥(我相信每个人类都有捡起一个棍状物就会中二地当剑一样挥来挥去的冲动 ww),还去了古城,但我对古文化提不起兴趣就权当晒太阳散步啦。

一些路途的记录(手机简单后期):

《一眼看过去是日出或傍晚但其实~》

《清冷的明月与艳丽的红花》

《去月球》

明天要去另一个城市大开吃戒!虽然这三天已经吃了两顿火锅和三顿年夜饭规格的大餐了

今年是第二次外出旅行而不在家过年啦,感觉很开心。在家 or 在外过年对我影响并不大,我对于家庭、团圆、新年一系列的词语没有什么感触,但是我很珍惜文化与习俗给了我这样「我说了算」式旅游的机会 ww

乐!!!!!!!!!!!