以前想到一个语境词典, 然后有大佬在 openai translator 中实现了。 语境词典的原贴在这里 https://www.v2ex.com/t/919114
贴一下这个贴子的目的是用来证明自己的想法不是意想天开。
这几天刷到 tinyfool 的英语轻松读。他做了客户端,服务端。代码量太多了,还只支持 ios 。 而且当一个单词有多种意思的时候,他还做不到选择一个最准确的意思。
然后我脑袋瓜子一激灵。 为什么不做一个类似于沉浸式翻译的 chrome 插件呢, 这样子, 打开网页就能用了,全平台支持。
我把 prompt 写好了, 不过我不会写 chrome 插件,我只会出主意。 我还有一大堆主意。 哪位大神愿意去实现在这个贴子里跟一下贴,谢谢。
你是一位英语老师,你将帮助我加速阅读,为了加速我的阅读,减少查字典的时间, 请将我给你的英语中词频大于 3000 的单词的意思标注在返回英文原文的生词,比如 我给你的英语句子是 On the whole, Harry thought he was to be congratulated on his idea of hiding here. 你的回答将是: On the whole, Harry thought he was to be congratulated(祝贺) on his idea of hiding here. 请记住一定要原文全部输出。 不只是输出单词和解释
试了一下 问 gpt: Few ideas are more unshakable than the notion that the rich keep getting richer while ordinary folks fall ever further behind. The belief that capitalism is rigged to benefit the wealthy and punish the workers has shaped how millions view the world, whom they vote for and whom they shake their fists at. It has been a spur to political projects on both left and right, from the interventionism of Joe Biden to the populism of Donald Trump. But is it true?
Even as the suspicion of free markets has hardened, evidence for the argument that inequality is rising in the rich world has become flimsier. Wage gaps are shrinking. Since 2016 real weekly earnings for those at the bottom of America’s pay distribution have grown faster than those at the top. Since the covid-19 pandemic this wage compression has gone into overdrive; according to one estimate, it has been enough to reverse an extraordinary 40% of the pre-tax wage inequality that emerged during the previous 40 years. A blue-collar bonanza is under way. gpt 问答: Few ideas are more unshakable than the notion that the rich keep getting richer while ordinary folks fall ever further behind. The belief that capitalism is rigged to benefit(受益) the wealthy and punish the workers has shaped how millions view the world, whom they vote for and whom they shake their fists at. It has been a spur(刺激) to political projects on both left and right, from the interventionism(干涉主义) of Joe Biden to the populism(民粹主义) of Donald Trump. But is it true?
Even as the suspicion of free markets has hardened, evidence for the argument that inequality is rising in the rich world has become flimsier(不牢固的). Wage gaps are shrinking. Since 2016 real weekly earnings for those at the bottom of America’s pay distribution have grown faster than those at the top. Since the covid-19 pandemic this wage compression has gone into overdrive; according to one estimate, it has been enough to reverse an extraordinary 40% of the pre-tax wage inequality that emerged during the previous 40 years. A blue-collar bonanza(繁荣) is under way.
1
z7356995 OP 用 gpt 的回答, 里面包含有难词解释的英文。 来替换掉网页中不含解释的原文,以取消查词,加快阅读速度
|
2
thinkm 2023-12-02 10:57:12 +08:00
👍,有点东西,但不多
|
3
z7356995 OP 只是给会写插件的同学提供一个思路,我不会
|
4
sapjax 2023-12-02 13:38:56 +08:00 5
你这样全文输入和全文输出,token 太多,收费会很贵
还不如用我的扩展 https://github.com/sapjax/word-hunter 0 花费 开启中文翻译后,一定程度上和你要的效果差不多 不过有两个前提 1. 需要先把认识的单词标记掉 2. 单词有多个意思的时候,中文翻译不准确,好在鼠标悬浮会显示所有的意思,再不行单词词典窗口也有 chatgpt 的上下文翻译 |
5
wuligelovevc 2023-12-02 15:11:13 +08:00
我学英语的方法就好笨,只会背 T T
|
7
z7356995 OP @sapjax 我试了一下, 一次性把全文所有的生词单词都列出来。比一次次用 openai translator 查, 一次性全标注的 token 更少,上面那段话五个生词, 只用了 638 token. 而如果用语境词典的话,一个生词就要 400 左右。 单词在文中的准确度普通字典和 chatgpt 是没法比的
|
8
sapjax 2023-12-02 15:53:26 +08:00
@z7356995 #7
这么一段话,就需要 600 多个 token ,已经很多了。 另外还有一个问题,你如何确保给出翻译的单词是你不认识的单词? 词频大于 3000 这个条件太宽泛了,要准确的话,只能回到我的那个办法,就是排除法,排除认识的之后,剩下的就是不认识的,只不过这个排除的过程需要点时间,需要在日常使用时逐渐排除掉,会越用越准确。 |
9
lemonTreeTop 2023-12-02 16:01:20 +08:00
可以直接把不懂单词替换为中文来阅读,不需要括号更合适
|
13
z7356995 OP @sapjax 这个 3000 还是 5000 不就一个参数吗,搞个滑块让用户根据自己水平调整一下就好了,不认识的字准确性大概匹配就行,你懂了,再看一下解释也不浪费多大时间。对用户来说一年要是能读上 100 块钱的 token 英语资料,英语学习应该可以暴涨了
|
14
z7356995 OP 我这个主要是通过 gpt 标注,减少查词时间,增加单词在上下文中的准确度,减少阅读痛苦
|
15
VxJiahua 2023-12-02 17:03:23 +08:00 via Android
@sapjax 就是 openai-translator 的 OpenAI API Proxy 功能,或者是沉浸式翻译的自定义 API 接口地址
|
18
Archeb 2023-12-02 20:18:42 +08:00
做过一个类似的基于词频的翻译,不过是给 YouTube 视频字幕用的: https://greasyfork.org/en/scripts/460216
|
19
z7356995 OP @Archeb 谢谢,有没有 chatgpt 的版本,chatgpt 有上下文的关联,对一个单词有多种意思的时候会选择最正确的意思,比如 battery 有电池的意思,又有排炮的意思。他可以根据上下文选择最正确的意思
|