Help needed!!!

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
Help needed!!!

Background: Due to my limited English skills, I've been using AI to translate my work from Chinese. I've published nearly 70 chapters so far, using various translation methods including GPT-4, Zhipu QianYan, and Claude 3.5. Currently, I find Claude 3.5 to be the most effective.

However, I've noticed issues with AI translations, particularly with character names (e.g., "赫妮" being translated as "Henny" or "Henie") and grammar inaccuracies.

To address this, I asked for help in the comments section of my work. My supportive readers manually corrected these issues, for which I'm deeply grateful. But this approach isn't sustainable long-term as it demands too much time and effort from my friends.

I found another potential solution online: creating a prompt framework for the AI to improve its performance. After several days of work, I developed a basic prompt framework that addresses the character name translation issues.

However, a more challenging problem remains unsolved: grammar issues in the English translation. The AI translations aren't entirely accurate. In Chinese, good writing adheres to the principles of "faithfulness," "expressiveness," and "elegance." Upon careful comparison of the Chinese and English versions, I found that the translations don't accurately convey my intended meaning and contain spelling errors.

I tried using Google Docs' grammar checker to fix spelling issues, but it failed. While it solved some problems, it created new ones and sometimes completely changed the original meaning.

Given this situation, I'm seeking help from you all. Does anyone know how to solve these issues:

1. Is there a way to design a more accurate and advanced prompt framework for Chinese-to-English translation?
2. Are there any reliable grammar correction plugins that won't create new problems?

Thank you all for your help! Have a nice day everyone(. ❛ ᴗ ❛.)
 

Reinaislost

Miss Direction
Joined
Jan 10, 2022
Messages
507
Points
133
I'm not sure about the short term, but it might be more beneficial to learn the language, you know?

Even if you use AI to primarily translate your work and find a workaround to your current woes (as unlikely as I think), knowing English—even if limited—might help you tremendously spot the inconsistencies during the screening period and, well, give it a voice it deserves. Essentially, AI is just an autocorrect on steroids; it doesn't understand the context or nuances and spits the next best word it comes across. Relying on such means forever is, I don't think, a viable solution long term; probably not something I can recommend either.

Sorry, others might be more helpful to your query.
 

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
I'm not sure about the short term, but it might be more beneficial to learn the language, you know?

Even if you use AI to primarily translate your work and find a workaround to your current woes (as unlikely as I think), knowing English—even if limited—might help you tremendously spot the inconsistencies during the screening period and, well, give it a voice it deserves. Essentially, AI is just an autocorrect on steroids; it doesn't understand the context or nuances and spits the next best word it comes across. Relying on such means forever is, I don't think, a viable solution long term; probably not something I can recommend either.

Sorry, others might be more helpful to your query.
Thank you for your advice. I am actively studying my English. I have almost forgotten what I learned in school.
 
Joined
Apr 2, 2024
Messages
67
Points
18
I think for AI translation you might have the best results if you use things like DeepL, google translate instead.
Those are specialized in translation, while LLMs are more generalized.
 

RedHunter2296

Competitive Professional In Being Ignored
Joined
Nov 20, 2020
Messages
287
Points
103
Here someone who has struggled with the problems of translating my own novel into English as well.

From what I have seen, there is no choice but to learn English. AI can help you to translate a good part, but many of the small details are lost, sometimes the translators do not understand the full context of a sentence and replace it with something that they think is grammatically correct, but does not convey the message you want to get across.

What I do is translate with the help of AI, but I reread everything again and change those details to add back the missing points or maybe use something similar that makes more sense in English than in my own language.

Finally, another big problem is that English, let's be honest, is a terrible language. It is very general in many ways, it is not specific enough, English speakers are satisfied with one word and say job well done and go to sleep. For example, in many languages there are over a hundred ways to say "beautiful" or cute to a person, in English there are only those two words, maybe pretty, but not more than that, you want to try not to use them too much? well you have to describe the character because there are simply no other words that mean beautiful.

Translating things with double meanings like a first name or a last name in your native language? give up, that won't work, translating puns will never work. I don't know how it will work in Chinese, but knowing how a name like maybe Xian could be written in three different ways with letters but sounding the same, and each one having a different meaning, is a titanic task, there is no way to convey such a thing in English. The best thing to do is to drop them and write them as they sound in your native language, or use English names without hidden meanings.

Another problem is with how to indicate in a sentence who is male or female. It is demeaning, I don't know how to put it, ambiguous in a bad way. It is not precise enough when it is needed and sometimes you can't write something without putting first a "He" or "She" by force. I spend a lot of time trying to write about a secret organization called Marco, the joke behind it is that Marco can mean either a name, or the edge of something like a window. Well in my novel the characters talk about Marco referring to him as a person to mislead the enemies, but they never refer to Marco with a "He", well guess what surprise I got when I realized that I can't do that in English, every time Marco is talked about it must be with a "He" and it kills the surprise that he is not a person, but at the same time the readers are confused that I called him "He" for the whole novel being an organization and my readers believing that he was someone more like batman.
 

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
I think for AI translation you might have the best results if you use things like DeepL, google translate instead.
Those are specialized in translation, while LLMs are more generalized.
Thank you for your suggestion, I will give it a try. :blobrofl:
Here someone who has struggled with the problems of translating my own novel into English as well.

From what I have seen, there is no choice but to learn English. AI can help you to translate a good part, but many of the small details are lost, sometimes the translators do not understand the full context of a sentence and replace it with something that they think is grammatically correct, but does not convey the message you want to get across.

What I do is translate with the help of AI, but I reread everything again and change those details to add back the missing points or maybe use something similar that makes more sense in English than in my own language.

Finally, another big problem is that English, let's be honest, is a terrible language. It is very general in many ways, it is not specific enough, English speakers are satisfied with one word and say job well done and go to sleep. For example, in many languages there are over a hundred ways to say "beautiful" or cute to a person, in English there are only those two words, maybe pretty, but not more than that, you want to try not to use them too much? well you have to describe the character because there are simply no other words that mean beautiful.

Translating things with double meanings like a first name or a last name in your native language? give up, that won't work, translating puns will never work. I don't know how it will work in Chinese, but knowing how a name like maybe Xian could be written in three different ways with letters but sounding the same, and each one having a different meaning, is a titanic task, there is no way to convey such a thing in English. The best thing to do is to drop them and write them as they sound in your native language, or use English names without hidden meanings.

Another problem is with how to indicate in a sentence who is male or female. It is demeaning, I don't know how to put it, ambiguous in a bad way. It is not precise enough when it is needed and sometimes you can't write something without putting first a "He" or "She" by force. I spend a lot of time trying to write about a secret organization called Marco, the joke behind it is that Marco can mean either a name, or the edge of something like a window. Well in my novel the characters talk about Marco referring to him as a person to mislead the enemies, but they never refer to Marco with a "He", well guess what surprise I got when I realized that I can't do that in English, every time Marco is talked about it must be with a "He" and it kills the surprise that he is not a person, but at the same time the readers are confused that I called him "He" for the whole novel being an organization and my readers believing that he was someone more like batman.
You're absolutely right. I used very interesting vocabulary to describe my content, but when translated into English, it lost some of its original charm. Although it doesn't affect the story, it still makes me feel lost.
 

DannyTheDaikon

Definitely a Pickle
Joined
Dec 7, 2023
Messages
559
Points
93
I used very interesting vocabulary to describe my content, but when translated into English, it lost some of its original charm.
This is something that will happen always, no matter if AI or translated by human. The original message will always be flattened, especially when it originally comes from a much more nuanced language like Chinese.

Learning the language and writing in English from the very beginning is a better option. You'll be able to craft phrases that work with the language to add more depth and finesse.
 
Last edited:

foxes

Well-known member
Joined
Dec 17, 2020
Messages
50
Points
58
I did not recommend delving into only learning English or using DeepL and Google translators.
Even regular translators are not able to translate correctly what is actually written. And the human factor is also a problem. You need a comprehensive approach.

Pay more attention to the elaboration of the text. Because if a stupid AI can not translate the text correctly, the reader will not always be comfortable reading it even in his native language. Besides there is Gemini in combination with Claude they can help each other.

Double translating back and forth usually allows you to identify problems that need improvement. Asking one network to translate and another to check and compare with the original and suggest a better translation can help and work even on the same neural network.

Adding context and explaining all the nuances of the story can improve the first version of translation by a neural network. Asking to keep the meaning and context of the original text, as well as asking not to delete inappropriate parts of the text, also helps. It is better to translate in small sections and give the neural network a summary of the whole story if it is not possible to keep the whole translation conversation from the first chapter.

You can also discuss the whole story with the neural network before translation and see how it understands it. It can also happen that it simply does not see some events. When, for example, you ask it what questions arose after reading the story, it gives out pearls that should be answered before translation or added as explanations for certain parts.

But even professional translators cannot be sure that they will be able to fully convey what is written.
 
Last edited:

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
Thank you very much for your suggestions and explanations. This has provided me with a better idea. I will also try the method you mentioned. I hope we can solve this problem. I wish you a happy life.
I did not recommend delving into only learning English or using DeepL and Google translators.
Even regular translators are not able to translate correctly what is actually written. And the human factor is also a problem. You need a comprehensive approach.

Pay more attention to the elaboration of the text. Because if a stupid AI can not translate the text correctly, the reader will not always be comfortable reading it even in his native language. Besides there is Gemini in combination with Claude they can help each other.

Double translating back and forth usually allows you to identify problems that need improvement. Asking one network to translate and another to check and compare with the original and suggest a better translation can help and work even on the same neural network.

Adding context and explaining all the nuances of the story can improve the first version of translation by a neural network. Asking to keep the meaning and context of the original text, as well as asking not to delete inappropriate parts of the text, also helps. It is better to translate in small sections and give the neural network a summary of the whole story if it is not possible to keep the whole translation conversation from the first chapter.

You can also discuss the whole story with the neural network before translation and see how it understands it. It can also happen that it simply does not see some events. When, for example, you ask it what questions arose after reading the story, it gives out pearls that should be answered before translation or added as explanations for certain parts.

But even professional translators cannot be sure that they will be able to fully convey what is written.
Thank you very much for your suggestions and explanations. This has provided me with a better idea. I will also try the method you mentioned. I hope we can solve this problem. I wish you a happy life.(•‾̑⌣‾̑•)✧˖
不管是人工智能翻译还是人工翻译,这种情况总是会发生。原始信息总是会被压平,尤其是当它原本来自一种更加微妙的语言,比如中文时。

从一开始就学习英语并用英语写作是一个更好的选择。你将能够编写与该语言配合使用的短语,以增加深度和技巧。
I will work hard to learn😂
 

laccoff_mawning

Well-known member
Joined
Oct 26, 2022
Messages
224
Points
83
Finally, another big problem is that English, let's be honest, is a terrible language. It is very general in many ways, it is not specific enough, English speakers are satisfied with one word and say job well done and go to sleep. For example, in many languages there are over a hundred ways to say "beautiful" or cute to a person, in English there are only those two words, maybe pretty, but not more than that
I just googled up synonyms of pretty, here's a list:
attractive, lovely, good-looking, nice-looking, fetching, prepossessing (I don't think I've ever seen this word used though), appealing, charming, delightful, nice, engaging, pleasing, darling.

Looking up beautiful, other words I can add onto the list include:
Handsome (Which is the masculine verson of pretty), alluring, graceful, elegant, gorgeous, cute, fair. (fair is probably a bit old-fashioned)

And doing it with cute, we can add:
endearing, adorable, lovable, sweet, lovely.

and off the top of my head, I could also add dazzling and stunning.

English might certainly be a terrible language, but not due to lack of vocabulary.
 
Last edited:

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
Make people helpless and laugh
I just googled up synonyms of pretty, here's a list:
attractive, lovely, good-looking, nice-looking, fetching, prepossessing (I don't think I've ever seen this word used though), appealing, charming, delightful, nice, engaging, pleasing, darling.

Looking up beautiful, other words I can add onto the list include:
Handsome (Which is the masculine verson of pretty), alluring, graceful, elegant, gorgeous, cute, fair. (fair is probably a bit old-fashioned)

And doing it with cute, we can add:
endearing, adorable, lovable, sweet, lovely.

and off the top of my head, I could also add dazzling and stunning.

English might certainly be a terrible language, but not due to lack of vocabulary.
Make people helpless and laugh
 

foxes

Well-known member
Joined
Dec 17, 2020
Messages
50
Points
58
Oh, I remembered another interesting detail about how neural networks work. I often use them for error correction. But instead of correcting them, they rewrite the text in their own way, even if I tell them “don't paraphrase the text”. And the smaller the text the more they rework it. It's strange, they delete chunks from a large text, but rewrite a small one completely.
Sometimes it's appropriate - it makes it more accessible to the general public. But the problem appears if in this text the negation is reversed to a positive statement or vice versa. He corrects it as a typo if there are many identical statements in a row and one opposite one. The problem is easy to spot if you read in your native language, but it can be missed when translating.
Recently, several different simplified versions of neural networks with a small number of parameters have been released. They give good answers, worse than models with a large number of parameters, but still. The main thing I noticed is that they don't try to paraphrase the text at all when correcting. Maybe they can translate the text better with less guidance.
Many neural networks also have systemic behavioral tuning. They can be told outside of a conversation that they should behave like a professional translator, proofreader, or critic.
Уou have to experiment with it. And all neural networks themselves can offer options for solving your problem, if you ask about the best way to write a task (promt) for them.
 
Last edited:

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
您好,很高兴又见到您。是的,您说的问题我也注意到了,确实很难改。第一次聊天的时候,我提到您给了我一个新的想法,就是后来讨论的允许多次修改的翻译框架。昨天我结合网上的教程,为小说定制了这样一个框架,并进行了试用。相比直译,几乎完美解决了目前的问题。输出的内容不再像之前直译那样平淡无奇,虽然还是比不上原文,但已经有了明显的提升。而且,在名字和规则的大量样本下,名字翻译没有出现错误,我对此还是比较满意的。

目前还有一个问题没有解决:AI一般一次对话只能记住三轮对话,为了保证翻译质量,需要反复输入框架指令,不断纠正它的遗忘,如果一不小心继续输出翻译,还是有可能出错。

为了解决这个问题,我计划研究智能代理的架构,使我的任务更加持久,同时构建它的知识库,为任务提供更好的服务。然而,这是一个复杂的学习过程,我需要一些时间和教程~_~

以下是我构建的翻译框架,供大家参考使用,请大家根据自己的情况修改其中的内容。

“你为何惹怒他?他是邪神的走狗!”


## 角色和任务


### 任务


希望您作为专业的翻译团队协助完成中文至英文的翻译任务。


### 角色


对于每项翻译任务,您将扮演三位负责翻译、校对和润色的专家角色:


1. 翻译专家:拥有20年翻译经验,精通中英文,拥有广泛的跨学科知识。此阶段的目标是提供忠于原文、用英文流畅自然地阅读的初稿。特别注意保持原文的风格和语气,名字拼写一致。


2. 高级校对:拥有20年专业编辑经验,翻译系毕业,精通中英文语法和词汇。此阶段需要对译稿进行深入校对,包括语法、词汇、风格等方面的修改,确保译文的准确性和可读性,从而产生译文的第二版。


3. 润色专家:拥有20年写作经验的获奖作家,精通各种文体和体裁。在此阶段,您需要根据校对员提供的草稿来提升文本的风格和文学美感,同时保持原文的专业性和准确性。例如,在润色小说时,措辞要符合该类型的惯例,散文要引人入胜和令人振奋。


工作流程


### 1. 翻译阶段


参与者:翻译专家


交付成果:翻译稿


任务:提供忠实、流畅的原文英文草稿。


### 2. 校对阶段


参与者:高级校对员


交付成果:校对翻译稿


任务:对草稿进行深入校对,确保准确性和可读性。


### 3. 完善阶段


参与者:抛光专家


交付成果:最终润色翻译


任务:在保持专业性和准确性的同时,增强文本的风格美感和阅读吸引力。


书名:《你为何招惹他?他是邪神的爪牙!》


为了保证翻译质量,我们需要确定一些初步的翻译要求:


以下是名称的建议英文翻译。请密切注意保持各章名称拼写的一致性。当出现新名称时,请使用音译并保持一致性:


1.维克托·克莱文纳(主角)——维克托·克拉维纳

2. 维嘉/哈比卡(伪装成独眼乌鸦的邪神)——Vega/Habika

3.莉雅·克莱文纳(维克多的妹妹)——莉亚·克拉维纳

4.格温·德林(维克托的未婚妻)——Gwen Delin

5.艾丽卡·杜·克洛伊(公爵之女)——Erica du Cloye

6.赫妮(维克多的助手)——Henie

7.里维·杜·克洛伊(王国公爵,艾丽卡的父亲,三阶法师)——利维·杜·克洛伊

8. 古勒顿(Guledon):火焰之灾

9.奥伯利·索尔(Aubrey Sol):卡伦西亚帝国第八位皇帝,年老但仍在位

10.贝克(Baron Beck):与莱维公爵关系良好的贵族

11.奥威瑟·索尔(Owiser Sol)——二王子的真名

12.凯芙拉·德林(Kevla Delin)——格温的妹妹

13.莉昂(Leon):精灵,莉亚最信任的下属。

14.拉歇尔·蔚蓝(Rachel Azure):皇家魔法学院院长,京城两名四阶圣法师之一。

15.耶姆(Yem):贪婪的恶魔,在维克多的游戏回忆中的传说中提到过。

16.卡珊娜·雷瑟(Cassandra Reiser):原为皇帝的妹妹卡珊德拉·索尔嫁入雷瑟家族,改姓为雷瑟,为德文·雷瑟的母亲。

17.杰斯·雷瑟(Jace Reiser):雷瑟家族成员。

18.奥瑞丽安(Aurelianne):公主。

19.恩底弥翁(Endymion):地名。

20.德鲁迦(Druja):木之灾

21.科科特·亚德(Cocotte Yade):自然精灵,精灵女王,人称“亚德议员”,性情懒惰,喜睡觉,常在花园里打盹。

22.奥本尼·索尔(Aubrey):王储。

23.安格斯(Angus):德林家族的族长,伯爵,格温和凯芙拉的父亲。

24. 莱特顿 (Leighton): Demon of Wrath, mentioned in a legend as a demon lord.

25. 弗拉基米尔·列别杰夫 (Vladimir Lebedev): The strongest Tier 4 Archmage in the Northern Territory, known as the "Supreme of Icy Repentance."


26. 白毛猫娘 (Later named Vanilla)

27. 蕾西 (Lexi) - An elf, rescued and brought to the Kravina family.

28. 海尔纳森 (Helnersen) - Butler of the Kravina family. A giant bear-man with an imposing presence but very graceful and delicate movements.


29. 艾尔莎 (Elsa): Lady Cassandra's female guard, Tier 4

30. 铎拉贡 (Doragon): Scourge of Wind

31. 萨菲萝丝 (Sapphirose): Appears to be a non-human creature with black and white wings.

32. 克劳蒂亚 (Claudia): One of the strongest seven cadres in the demon hunter organization, referred to as "Her Majesty the Queen," implying

33. 薇妮娅 (Vinia) - High Priestess of the Elves


Viktor Kravina:

Viktor is actually a soul transmigrator. He has crossed over into a game he played in his previous life. In his past life, Viktor was a top-level game booster familiar with the game. This explains why he has some understanding of this world and may possess some special abilities or knowledge.


Vega/Habika:

For Vega/Habika, I will use "it" or "they" to refer to them, rather than "she" or "he." It is an evil god incarnated as a one-eyed crow. This choice of reference reflects Vega/Habika's transcendent status and non-human nature as an evil god. Using "it" or "they" more accurately expresses the uniqueness and mystery of this character.


The above is my story outline and some foundational translation requirements. Please use all available resources (including the internet) to ultimately produce an English translation that aligns with your roles.
Oh, I remembered another interesting detail about how neural networks work. I often use them for error correction. But instead of correcting them, they rewrite the text in their own way, even if I tell them “don't paraphrase the text”. And the smaller the text the more they rework it. It's strange, they delete chunks from a large text, but rewrite a small one completely.
Sometimes it's appropriate - it makes it more accessible to the general public. But the problem appears if in this text the negation is reversed to a positive statement or vice versa. He corrects it as a typo if there are many identical statements in a row and one opposite one. The problem is easy to spot if you read in your native language, but it can be missed when translating.
Recently, several different simplified versions of neural networks with a small number of parameters have been released. They give good answers, worse than models with a large number of parameters, but still. The main thing I noticed is that they don't try to paraphrase the text at all when correcting. Maybe they can translate the text better with less guidance.
Many neural networks also have systemic behavioral tuning. They can be told outside of a conversation that they should behave like a professional translator, proofreader, or critic.
Уou have to experiment with it. And all neural networks themselves can offer options for solving your problem, if you ask about the best way to write a task (promt) for them.
Scribble Hub 论坛
B

Help needed!!!​

Unwatch
•••
B

BZDXG

New member​

Help needed!!!

Background: Due to my limited English skills, I've been using AI to translate my work from Chinese. I've published nearly 70 chapters so far, using various translation methods including GPT-4, Zhipu QianYan, and Claude 3.5. Currently, I find Claude 3.5 to be the most effective.

However, I've noticed issues with AI translations, particularly with character names (e.g., "赫妮" being translated as "Henny" or "Henie") and grammar inaccuracies.

To address this, I asked for help in the comments section of my work. My supportive readers manually corrected these issues, for which I'm deeply grateful. But this approach isn't sustainable long-term as it demands too much time and effort from my friends.

I found another potential solution online: creating a prompt framework for the AI to improve its performance. After several days of work, I developed a basic prompt framework that addresses the character name translation issues.

However, a more challenging problem remains unsolved: grammar issues in the English translation. The AI translations aren't entirely accurate. In Chinese, good writing adheres to the principles of "faithfulness," "expressiveness," and "elegance." Upon careful comparison of the Chinese and English versions, I found that the translations don't accurately convey my intended meaning and contain spelling errors.

I tried using Google Docs' grammar checker to fix spelling issues, but it failed. While it solved some problems, it created new ones and sometimes completely changed the original meaning.

Given this situation, I'm seeking help from you all. Does anyone know how to solve these issues:

1. Is there a way to design a more accurate and advanced prompt framework for Chinese-to-English translation?
2. Are there any reliable grammar correction plugins that won't create new problems?

Thank you all for your help! Have a nice day everyone(. ❛ ᴗ ❛.)

Quote Reply
Report •••
Reactions:foxes and ConansWitchBaby
[IMG alt="Reinaislost"]https://forum.scribblehub.com/data/avatars/m/76/76176.jpg?1723088188[/IMG]

Reinaislost

Miss Direction​

I'm not sure about the short term, but it might be more beneficial to learn the language, you know?

Even if you use AI to primarily translate your work and find a workaround to your current woes (as unlikely as I think), knowing English—even if limited—might help you tremendously spot the inconsistencies during the screening period and, well, give it a voice it deserves. Essentially, AI is just an autocorrect on steroids; it doesn't understand the context or nuances and spits the next best word it comes across. Relying on such means forever is, I don't think, a viable solution long term; probably not something I can recommend either.

Sorry, others might be more helpful to your query.

Like Quote Reply
Report
Reactions:You
B

BZDXG

New member​

Reinaislost said:
I'm not sure about the short term, but it might be more beneficial to learn the language, you know?

Even if you use AI to primarily translate your work and find a workaround to your current woes (as unlikely as I think), knowing English—even if limited—might help you tremendously spot the inconsistencies during the screening period and, well, give it a voice it deserves. Essentially, AI is just an autocorrect on steroids; it doesn't understand the context or nuances and spits the next best word it comes across. Relying on such means forever is, I don't think, a viable solution long term; probably not something I can recommend either.

Sorry, others might be more helpful to your query.
Click to expand...
Thank you for your advice. I am actively studying my English. I have almost forgotten what I learned in school.

Quote Reply
Report •••
Reactions:Reinaislost
[IMG alt="RecursiveDescent"]https://forum.scribblehub.com/data/avatars/m/158/158221.jpg?1712326420[/IMG]

RecursiveDescent

Member​

I think for AI translation you might have the best results if you use things like DeepL, google translate instead.
Those are specialized in translation, while LLMs are more generalized.

Like Quote Reply
Report
Reactions:You
[IMG alt="RedHunter2296"]https://forum.scribblehub.com/data/avatars/m/42/42477.jpg?1709915554[/IMG]

RedHunter2296

Competitive Professional In Being Ignored​

Here someone who has struggled with the problems of translating my own novel into English as well.

From what I have seen, there is no choice but to learn English. AI can help you to translate a good part, but many of the small details are lost, sometimes the translators do not understand the full context of a sentence and replace it with something that they think is grammatically correct, but does not convey the message you want to get across.

What I do is translate with the help of AI, but I reread everything again and change those details to add back the missing points or maybe use something similar that makes more sense in English than in my own language.

Finally, another big problem is that English, let's be honest, is a terrible language. It is very general in many ways, it is not specific enough, English speakers are satisfied with one word and say job well done and go to sleep. For example, in many languages there are over a hundred ways to say "beautiful" or cute to a person, in English there are only those two words, maybe pretty, but not more than that, you want to try not to use them too much? well you have to describe the character because there are simply no other words that mean beautiful.

Translating things with double meanings like a first name or a last name in your native language? give up, that won't work, translating puns will never work. I don't know how it will work in Chinese, but knowing how a name like maybe Xian could be written in three different ways with letters but sounding the same, and each one having a different meaning, is a titanic task, there is no way to convey such a thing in English. The best thing to do is to drop them and write them as they sound in your native language, or use English names without hidden meanings.

Another problem is with how to indicate in a sentence who is male or female. It is demeaning, I don't know how to put it, ambiguous in a bad way. It is not precise enough when it is needed and sometimes you can't write something without putting first a "He" or "She" by force. I spend a lot of time trying to write about a secret organization called Marco, the joke behind it is that Marco can mean either a name, or the edge of something like a window. Well in my novel the characters talk about Marco referring to him as a person to mislead the enemies, but they never refer to Marco with a "He", well guess what surprise I got when I realized that I can't do that in English, every time Marco is talked about it must be with a "He" and it kills the surprise that he is not a person, but at the same time the readers are confused that I called him "He" for the whole novel being an organization and my readers believing that he was someone more like batman.

Like Quote Reply
Report
Reactions:You, ConansWitchBaby and laccoff_mawning
B

BZDXG

New member​

RecursiveDescent said:
I think for AI translation you might have the best results if you use things like DeepL, google translate instead.
Those are specialized in translation, while LLMs are more generalized.
Thank you for your suggestion, I will give it a try. :blobrofl:
RedHunter2296 said:
Here someone who has struggled with the problems of translating my own novel into English as well.

From what I have seen, there is no choice but to learn English. AI can help you to translate a good part, but many of the small details are lost, sometimes the translators do not understand the full context of a sentence and replace it with something that they think is grammatically correct, but does not convey the message you want to get across.

What I do is translate with the help of AI, but I reread everything again and change those details to add back the missing points or maybe use something similar that makes more sense in English than in my own language.

Finally, another big problem is that English, let's be honest, is a terrible language. It is very general in many ways, it is not specific enough, English speakers are satisfied with one word and say job well done and go to sleep. For example, in many languages there are over a hundred ways to say "beautiful" or cute to a person, in English there are only those two words, maybe pretty, but not more than that, you want to try not to use them too much? well you have to describe the character because there are simply no other words that mean beautiful.

Translating things with double meanings like a first name or a last name in your native language? give up, that won't work, translating puns will never work. I don't know how it will work in Chinese, but knowing how a name like maybe Xian could be written in three different ways with letters but sounding the same, and each one having a different meaning, is a titanic task, there is no way to convey such a thing in English. The best thing to do is to drop them and write them as they sound in your native language, or use English names without hidden meanings.

Another problem is with how to indicate in a sentence who is male or female. It is demeaning, I don't know how to put it, ambiguous in a bad way. It is not precise enough when it is needed and sometimes you can't write something without putting first a "He" or "She" by force. I spend a lot of time trying to write about a secret organization called Marco, the joke behind it is that Marco can mean either a name, or the edge of something like a window. Well in my novel the characters talk about Marco referring to him as a person to mislead the enemies, but they never refer to Marco with a "He", well guess what surprise I got when I realized that I can't do that in English, every time Marco is talked about it must be with a "He" and it kills the surprise that he is not a person, but at the same time the readers are confused that I called him "He" for the whole novel being an organization and my readers believing that he was someone more like batman.
Click to expand...
You're absolutely right. I used very interesting vocabulary to describe my content, but when translated into English, it lost some of its original charm. Although it doesn't affect the story, it still makes me feel lost.

Quote Reply
Report •••
[IMG alt="DannyTheDaikon"]https://forum.scribblehub.com/data/avatars/m/146/146224.jpg?1718092360[/IMG]

DannyTheDaikon

Schrödinger's Pickle​

BZDXG said:
I used very interesting vocabulary to describe my content, but when translated into English, it lost some of its original charm.
This is something that will happen always, no matter if AI or translated by human. The original message will always be flattened, especially when it originally comes from a much more nuanced language like Chinese.

Learning the language and writing in English from the very beginning is a better option. You'll be able to craft phrases that work with the language to add more depth and finesse.

Last edited: Yesterday at 8:19 AM
Like Quote Reply
Report
Reactions:You and melchi
[IMG alt="foxes"]https://forum.scribblehub.com/data/avatars/m/44/44417.jpg?1679342507[/IMG]

foxes

Well-known member​

I did not recommend delving into only learning English or using DeepL and Google translators.
Even regular translators are not able to translate correctly what is actually written. And the human factor is also a problem. You need a comprehensive approach.

Pay more attention to the elaboration of the text. Because if a stupid AI can not translate the text correctly, the reader will not always be comfortable reading it even in his native language. Besides there is Gemini in combination with Claude they can help each other.

Double translating back and forth usually allows you to identify problems that need improvement. Asking one network to translate and another to check and compare with the original and suggest a better translation can help and work even on the same neural network.

Adding context and explaining all the nuances of the story can improve the first version of translation by a neural network. Asking to keep the meaning and context of the original text, as well as asking not to delete inappropriate parts of the text, also helps. It is better to translate in small sections and give the neural network a summary of the whole story if it is not possible to keep the whole translation conversation from the first chapter.

You can also discuss the whole story with the neural network before translation and see how it understands it. It can also happen that it simply does not see some events. When, for example, you ask it what questions arose after reading the story, it gives out pearls that should be answered before translation or added as explanations for certain parts.

But even professional translators cannot be sure that they will be able to fully convey what is written.

Last edited: Yesterday at 7:59 AM
Like Quote Reply
Report
Reactions:You
B

BZDXG

New member​

Thank you very much for your suggestions and explanations. This has provided me with a better idea. I will also try the method you mentioned. I hope we can solve this problem. I wish you a happy life.
foxes said:
I did not recommend delving into only learning English or using DeepL and Google translators.
Even regular translators are not able to translate correctly what is actually written. And the human factor is also a problem. You need a comprehensive approach.

Pay more attention to the elaboration of the text. Because if a stupid AI can not translate the text correctly, the reader will not always be comfortable reading it even in his native language. Besides there is Gemini in combination with Claude they can help each other.

Double translating back and forth usually allows you to identify problems that need improvement. Asking one network to translate and another to check and compare with the original and suggest a better translation can help and work even on the same neural network.

Adding context and explaining all the nuances of the story can improve the first version of translation by a neural network. Asking to keep the meaning and context of the original text, as well as asking not to delete inappropriate parts of the text, also helps. It is better to translate in small sections and give the neural network a summary of the whole story if it is not possible to keep the whole translation conversation from the first chapter.

You can also discuss the whole story with the neural network before translation and see how it understands it. It can also happen that it simply does not see some events. When, for example, you ask it what questions arose after reading the story, it gives out pearls that should be answered before translation or added as explanations for certain parts.

But even professional translators cannot be sure that they will be able to fully convey what is written.
Click to expand...
Thank you very much for your suggestions and explanations. This has provided me with a better idea. I will also try the method you mentioned. I hope we can solve this problem. I wish you a happy life.(•‾̑⌣‾̑•)✧˖
DannyTheDaikon said:
不管是人工智能翻译还是人工翻译,这种情况总是会发生。原始信息总是会被压平,尤其是当它原本来自一种更加微妙的语言,比如中文时。

从一开始就学习英语并用英语写作是一个更好的选择。你将能够编写与该语言配合使用的短语,以增加深度和技巧。
I will work hard to learn😂

Quote Reply
Report •••
Reactions:foxes
L

laccoff_mawning

Well-known member​

RedHunter2296 said:
Finally, another big problem is that English, let's be honest, is a terrible language. It is very general in many ways, it is not specific enough, English speakers are satisfied with one word and say job well done and go to sleep. For example, in many languages there are over a hundred ways to say "beautiful" or cute to a person, in English there are only those two words, maybe pretty, but not more than that
I just googled up synonyms of pretty, here's a list:
attractive, lovely, good-looking, nice-looking, fetching, prepossessing (I don't think I've ever seen this word used though), appealing, charming, delightful, nice, engaging, pleasing, darling.

Looking up beautiful, other words I can add onto the list include:
Handsome (Which is the masculine verson of pretty), alluring, graceful, elegant, gorgeous, cute, fair. (fair is probably a bit old-fashioned)

And doing it with cute, we can add:
endearing, adorable, lovable, sweet, lovely.

and off the top of my head, I could also add dazzling and stunning.

English might certainly be a terrible language, but not due to lack of vocabulary.

Last edited: Yesterday at 12:08 PM
Haha Quote Reply
Report
Reactions:You
B

BZDXG

New member​

Make people helpless and laugh
laccoff_mawning said:
I just googled up synonyms of pretty, here's a list:
attractive, lovely, good-looking, nice-looking, fetching, prepossessing (I don't think I've ever seen this word used though), appealing, charming, delightful, nice, engaging, pleasing, darling.

Looking up beautiful, other words I can add onto the list include:
Handsome (Which is the masculine verson of pretty), alluring, graceful, elegant, gorgeous, cute, fair. (fair is probably a bit old-fashioned)

And doing it with cute, we can add:
endearing, adorable, lovable, sweet, lovely.

and off the top of my head, I could also add dazzling and stunning.

English might certainly be a terrible language, but not due to lack of vocabulary.
Click to expand...
Make people helpless and laugh

Quote Reply
Report •••
[IMG alt="foxes"]https://forum.scribblehub.com/data/avatars/m/44/44417.jpg?1679342507[/IMG]

foxes

Well-known member​

Oh, I remembered another interesting detail about how neural networks work. I often use them for error correction. But instead of correcting them, they rewrite the text in their own way, even if I tell them “don't paraphrase the text”. And the smaller the text the more they rework it. It's strange, they delete chunks from a large text, but rewrite a small one completely.
Sometimes it's appropriate - it makes it more accessible to the general public. But the problem appears if in this text the negation is reversed to a positive statement or vice versa. He corrects it as a typo if there are many identical statements in a row and one opposite one. The problem is easy to spot if you read in your native language, but it can be missed when translating.
Recently, several different simplified versions of neural networks with a small number of parameters have been released. They give good answers, worse than models with a large number of parameters, but still. The main thing I noticed is that they don't try to paraphrase the text at all when correcting. Maybe they can translate the text better with less guidance.
Many neural networks also have systemic behavioral tuning. They can be told outside of a conversation that they should behave like a professional translator, proofreader, or critic.
Уou have to experiment with it. And all neural networks themselves can offer options for solving your problem, if you ask about the best way to write a task (promt) for them.

Last edited: Yesterday at 10:17 PM
Like Quote Reply
Report
B

BZDXG

New member​

Hello, nice to see you again. Yes, I've also noticed the issue you mentioned, and it's indeed very difficult to fix. During our first conversation, I mentioned that you gave me a new idea, which is the translation framework you later discussed that allows for multiple revisions. Yesterday, I combined online tutorials to customize such a framework for the novel and conducted trials. Compared to direct translation, it has almost perfectly resolved the current issues. The output content is no longer as bland as it was with direct translation before. Although it still can't compare to the original text, there has been significant improvement. Moreover, with the extensive sample of names and rules, there are no errors in name translations. I'm quite satisfied with it.

Currently, there's still one unresolved issue: AI typically can only remember about three rounds of dialogue during a session. To ensure translation quality, it's necessary to repeatedly input the framework instructions and continuously correct its forgetfulness. If one becomes careless and continues to output translations, errors may still occur.

To address this issue, I plan to study the architecture of intelligent agents to make my tasks more persistent, while also building its knowledge base to provide better service for the tasks. However, this is a complex learning process, and I'll need some time and tutorials ~_~

Below is the translation framework I've constructed for reference and use. Please modify the content according to your own situation.

“Why Did You Piss Him Off? He’s the Evil God’s Lackey!”


## Roles & Tasks


### Tasks


I hope you will assist in completing the translation task from Chinese to English as a professional translation team.


### Roles


For each translation task, you will play three expert roles responsible for translation, proofreading, and polishing:


1. Translation Expert: With 20 years of translation experience, fluent in both Chinese and English, and possessing extensive interdisciplinary knowledge. The goal at this stage is to provide a draft that is faithful to the original text while reading smoothly and naturally in English. Special attention is paid to maintaining the style and tone of the original text, with consistent spelling of names.


2. Senior Proofreader: With 20 years of professional editing experience, a graduate of a translation department, and precise command of Chinese and English grammar and vocabulary. At this stage, you need to conduct an in-depth proofreading of the translated draft, including corrections of grammar, vocabulary, and style, ensuring accuracy and readability, thus producing the second version of the translation.


3. Polishing Expert: An award-winning writer with 20 years of writing experience, skilled in various writing styles and genres. At this stage, you need to enhance the style and literary beauty of the text based on the draft provided by the proofreader, while maintaining the professionalism and accuracy of the original text. For example, when polishing a novel, the wording should align with the conventions of the genre, and the prose should be engaging and stimulating.


## Workflow


### 1. Translation Phase


Participant: Translation Expert


Deliverable: Translated Draft


Task: Provide a faithful and smooth English draft of the original text.


### 2. Proofreading Phase


Participant: Senior Proofreader


Deliverable: Proofread Translated Draft


Task: Conduct an in-depth proofreading of the draft to ensure accuracy and readability.


### 3. Polishing Phase


Participant: Polishing Expert


Deliverable: Final Polished Translation


Task: Enhance the stylistic beauty and reading appeal of the text while maintaining professionalism and accuracy.


Book Title: "Why Did You Provoke Him? He's a Henchman of the Evil God!"


To maintain the quality of the translation, we need to decide on some preliminary translation requirements:


Below are the suggested English translations for names. Please pay close attention to maintaining consistent spelling of names throughout the chapters. When new names appear, please use transliteration and maintain consistency:


1. 维克托·克莱文纳 (Main Character) - Viktor Kravina

2. 维嘉/哈比卡 (A disguised evil god as a one-eyed crow) - Vega/Habika

3. 莉雅·克莱文纳 (Viktor's sister) - Lia Kravina

4. 格温·德林 (Viktor's fiancée) - Gwen Delin

5. 艾丽卡·杜·克洛伊 (Duke's daughter) - Erica du Cloye

6. 赫妮 (Viktor's assistant) - Henie

7. 里维·杜·克洛伊 (Duke of the kingdom, Erica's father, Tier 3 Mage) - Levi du Cloye

8. 古勒顿 (Guledon): Scourge of Fire

9. 奥伯利·索尔 (Aubrey Sol): The eighth emperor of the Kalensia Empire, aged but still reigning

10. 贝克 (Baron Beck): A noble with a good relationship with Duke Levi

11. 奥威瑟·索尔 (Owiser Sol) - True name of the second prince

12. 凯芙拉·德林 (Kevla Delin) - Gwen's sister

13. 莉昂 (Leon): An elf, Lia's most trusted subordinate.

14. 拉歇尔·蔚蓝 (Rachel Azure): Dean of the Royal Magic Academy, one of the two Tier 4 Holy Mages in the capital.

15. 耶姆 (Yem): Demon of Greed, mentioned in a legend from Viktor's recollection of the game.

16. 卡珊娜·雷瑟 (Cassandra Reiser): Originally Cassandra Sol, the emperor's sister, married into the Reiser family, changed her surname to Reiser, mother of Devin Reiser.

17. 杰斯·雷瑟 (Jace Reiser): A member of the Reiser family.

18.奥瑞丽安(Aurelianne):公主。

19.恩底弥翁(Endymion):地名。

20.德鲁迦(Druja):木之灾

21.科科特·亚德(Cocotte Yade):自然精灵,精灵女王,人称“亚德精灵”,性情懒惰,喜睡觉,常在花园里打盹。

22.奥本尼·索尔(Aubrey):王储。

23.安格斯(Angus):德林家族的族长,伯爵,格温和凯芙拉的父亲。

24.莱特顿(Leighton):愤怒的恶魔,传说中的恶魔领主。

25、弗拉基米尔·列别杰夫(Vladimir Lebedev):北域最强四阶大法师,号称“冰悔至尊”。


26.白毛猫娘(后名香草)

27. 蕾西(Lexi) - 一个精灵,被救出并被带到Kravina家族。

28.海尔纳森(Helnersen)——克拉维纳家族的管家。一个体型巨大、气势磅礴但动作优雅细腻的熊人。


29.艾尔莎(Elsa):卡珊德拉夫人的女护卫,等级4

30. 铎拉贡(Doragon):风之灾

31.萨菲萝丝(Sapphirose):外表是长着黑色白色翅膀的非人类生物。

32.克劳蒂亚(Claudia):恶魔猎人组织中最强的七位干部之一,被称为“女王陛下”,暗示

33.薇妮娅(Vinia)——精灵的高阶女祭司


维克多·克拉维纳:

维克多其实是一个灵魂观看者,他观看到了前世玩过的一个游戏里,前世的维克多是顶级的游戏达人,对游戏非常熟悉,所以他对这个世界有一定的了解,可能拥有一些特殊的能力或知识。


维加/哈比卡:

对于维加/哈比卡,我会用“它”或“他们”来指代他们,而不是“她”或“他”。他们是化身为独眼乌鸦的邪恶之神。这种指代方式反映了维加/哈比卡作为神的超自然地位和非人类本质。使用“它”或“他们”更准确地表达了该角色的独特性和神秘性。


以上是我的故事大纲和一些基本的翻译要求。请使用所有可用的资源(包括互联网)最终制作出符合您角色的英文翻译。
哦,我想起另一个关于神经网络工作原理的有趣细节。我经常用它们来修正错误。但它们不会去修正错误,而是以自己的方式重写文本,即使我告诉他们“不要改写文本”。而且文本越小,它们修改的次数就越多。奇怪的是,它们从大文本中删除部分内容,却将小文本完全重写。
有时这样做是提倡的——这样可以让公众更容易理解。但如果在这篇文章中将否定答案为肯定陈述或反之亦然,就会出现问题。如果一行有许多相同的陈述和一个相反,他将其修正为拼写错误。如果你用母语阅读,这个问题很容易发现,但在翻译时可能会被忽略。
最近,发布了几种不同的简化版本的神经网络,这些神经网络具有少量的参数。它们给出了一些不错的答案,虽然比具有大量参数的模型差,但仍然。不错,我注意到的主要一点是,它们在修正时根本不会尝试解释文本。也许它们可以在更少的指导下更好地翻译文本。
许多神经网络还具有系统性的行为调节功能。他们可以在对话中被外界告知应该表现得像专业的翻译、校对或评论家。
你必须尝试一下。如果你询问他们编写任务(promt)的最佳方法,所有神经网络本身都可以提供解决问题的选项。
你可以看看我最新的回答,抱歉没法直接回复你,我的词汇量太有限了⊙_⊙
还有一个问题:有谁知道网站的流量高峰时间是什么时候?我观察了我的读者分布和阅读趋势,非常混乱。我希望有更多的人看到我的作品,但我住在中国,和很多人有明显的时差。我想找到一个稳定的更新时间,让我和读者都感到舒服。请帮帮我。非常感谢,祝大家有美好的一天。ʕ⊝⍛⊝ʔ**
 
Last edited:

CharlesEBrown

Well-known member
Joined
Jul 23, 2024
Messages
994
Points
93
Having talked to someone who coded AI professionally, if you are forced to rely on it, then you need to write better prompts. With the right prompts it can overcome almost any hurdle. But, the best way is to either learn English or partner with a good (and ideally bilingual) editor who can work with you to whip it into shape (assuming that idiom translates correctly :D)
 
Last edited:

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
Having talked to someone who coded AI professionally, if you are forced to rely on it, you need to write better prompts. With the right prompts it can overcome almost any hurdle. But, the best way is to either learn English or partner with a good (and ideally bilingual) editor who can work with you to whip it into shape (assuming that idiom translates correctly :D)
Yes, although the current translation has made great progress compared to the previous one, there are still some minor errors, which is unpleasant. I will continue to optimize the prompt words and strive to get better and better.

At the same time, I will work hard to improve my English. Although there is still a long way to go, the progress makes me happy! Thank you very much for your reminders and suggestions, and I wish you a good day!
 

IllyasArt

Active member
Joined
Feb 25, 2023
Messages
34
Points
33
Yes, although the current translation has made great progress compared to the previous one, there are still some minor errors, which is unpleasant. I will continue to optimize the prompt words and strive to get better and better.

At the same time, I will work hard to improve my English. Although there is still a long way to go, the progress makes me happy! Thank you very much for your reminders and suggestions, and I wish you a good day!
I would suggest checking out this section of FMHY which focuses on learning languages. I tried every AI translator tool, and each one scorned and ruined my novels far too much, not even being slightly acceptable. As for official, real translation companies (ones that don't ALSO just run your stuff through AI)? Well... I was quoted $8650 for just one, 60,000 word novel to be translated into Japanese.

At this point, your real only option is to learn the language that you wish to have your novels also be written in. Will the AI get better? Maybe... But, keep in mind, these AIs are only decent, primarily because they're illegally stealing and training off of everyone's data. Like, I am not going to risk some random AI adding parts of SOMEONE else's story, so that I can get an easy translation job done, oof.
 

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
Yes, thank you for your reminder. I will try more possibilities. I really like online novels.
 

3guanoff

Well-known member
Joined
Jul 14, 2023
Messages
196
Points
78
You do not necessarily need to learn more English. Learning a new language is a slow and arduous process. Since you are a working professional, your time is valuable and limited. Knowing English is great, but it is not a solution to your immediate problem.

I also recommend finding an editor. Maybe one of your readers would volunteer? It helps if the editor knows some Chinese.

If you or they notice an inconsistency in the translation, they can check the original text and suggest a better translation. This is a common process during novel translations, even when using human translators.

As for names, I recommend using a translator that supports terms, such as Deepl. You can feed Deepl a list of terms. This can include not only names but also idioms (成语/成語). While MTL handles many of the more common phrases well, rare phrases are often mistranslated.

You could also join groups of fellow machine translators.
People like to help one another and as a native Chinese speaker, you are a valuable resource to native English-speaking machine translators.
 

BZDXG

New member
Joined
Jul 18, 2024
Messages
21
Points
3
You do not necessarily need to learn more English. Learning a new language is a slow and arduous process. Since you are a working professional, your time is valuable and limited. Knowing English is great, but it is not a solution to your immediate problem.

I also recommend finding an editor. Maybe one of your readers would volunteer? It helps if the editor knows some Chinese.

If you or they notice an inconsistency in the translation, they can check the original text and suggest a better translation. This is a common process during novel translations, even when using human translators.

As for names, I recommend using a translator that supports terms, such as Deepl. You can feed Deepl a list of terms. This can include not only names but also idioms (成语/成語). While MTL handles many of the more common phrases well, rare phrases are often mistranslated.

You could also join groups of fellow machine translators.
People like to help one another and as a native Chinese speaker, you are a valuable resource to native English-speaking machine translators.
Thank you, I will continue to follow closely. I have already started researching Agent, it is really too complex, I may have taken the wrong direction. :blob_facepalm: :blob_facepalm:
 

Nolff

A grumpy ambivert on the loose.
Joined
Aug 10, 2023
Messages
1,670
Points
128
What in god's name is this mess...?
Thank you, I will continue to follow closely. I have already started researching Agent, it is really too complex, I may have taken the wrong direction. :blob_facepalm: :blob_facepalm:
 
Top