The reason why there was considerable fluctuation in notation in the text I wrote this month was.(Text proofread along with ChatGPT)
The article we published this month had a level of notational fluctuation that I think was confusing to readers. Some sentences had unnaturally neat endings, or the first person was “I” instead of “me. There were also sentences that were structured in a way that made us wonder if they were intended to be reprinted in other media.
I think there were some things here and there in the text that were clearly different from the writing style I have cultivated for more than a quarter of a century. This is because I used ChatGPT to proofread the text (I proofread the text along with ChatGPT on why there was a considerable amount of notational fluctuation in the text I wrote this month.)The article we published this month had a level of notational fluctuation that I believe was unsettling to readers. Some sentences had unnaturally neat endings, or the first person was “I” instead of “me. There were also sentences structured in a way that made us wonder if they were intended to be reprinted in other media. I think there were some things here and there in the text that were clearly different from the writing style I have cultivated for more than a quarter of a century. This is because I used ChatGPT to proofread the text.
Of course, I did not create the sentences from scratch, but had ChatGPT proofread my previous sentences. However, simply telling ChatGPT to “proofread the following sentences” may not be enough to get it to proofread them properly. Therefore, I have done some prompt engineering, but I have not yet perfected it.
I think the free version of ChatGPT as of April 2023 is in a state of “you don’t know what you’re going to get until you open it, a ball and chain box”. Even if you give the AI the same instructions, the same constraints, and the same phrases, the answers that come back each time may be different. Sometimes the answer can look like it was written by a complete stranger.
The security aspect is also very questionable. As my personal experience, it is strongly recommended not to put any personal information on ChatGPT, because the risk of “others seeing the information” is much higher than the risk of AI learning.
At least for me, I do not want you to rewrite “I” as “I” when proofreading, and I do not want you to rewrite the endings in “is/masu” style, and I have written that in the constraints. However, at the GPT 3.5 level, this may be ignored. This is probably because my Japanese reading comprehension is still low. But it is not consistent, because sometimes it wakes up and works as it should.
Maybe it would work better if it was a $20/month GPT4. But if it doesn’t work, it’s a waste of money, so I have no plans to use the paid service at this time.
The ChatGPT prompt has been reworked many times and is already at version 10. The AI is very inconsistent, so it is difficult to get through the trial and error process. I am still trying to figure out if I should shorten the sentences to be proofread or reduce the number of constraints.
Basically, the current ChatGPT does not seem to be good at responding to prohibitions that say “don’t ~”. It seems that it would be better to steer the AI responses towards “do” and narrow the scope so that they are not misguided. However, even knowing this, it is still difficult to get a Japanese response with a Japanese command.
The reason I insist on proofreading scripts to this extent, spending many hours, more than 10 hours in fact, is that they sometimes return a sentence that makes me huff and puff.
My sentences are often verbose and wordy. However, when I let ChatGPT proofread my work, it sometimes rewrites my sentences to make them very readable by changing word order and conjunctions. I am surprised and impressed when I see how well it does this.
So we have further tuned ChatGPT to prevent the AI from summarizing. However, the free version of ChatGPT has a limit on the number of characters that can be entered, so it is necessary to split an article and have it read. Therefore, the level of proofreading of the first half of the article may be different from that of the second half.
Unfortunately, the AI cannot rewrite my writing perfectly. The reason for my poor writing is that I leave things as they are and publish them on the web without ever rereading them. Using ChatGPT makes little sense to me because I have to rewrite my writing many times.
Therefore, from now on, I will publish the text as it is proofread in such a snooty way. Since it is currently impossible to achieve the proofreading I was aiming for, I will stop this practice for the time being.
In the future, I foresee a more practical AI text editing feature. If such a feature is available, I would be willing to pay for it.
There are many “text editing tools” available today, but they are not very exciting to me, such as correcting misspelled words and forbidden terms. Of course, these features are useful, but they do not automatically correct them for me; I have to do it myself in the end. Also, when I load my text into such a tool, it often points out things like “missing particles”, which is a pain to deal with, so I don’t use it.
We can also ask ChatGPT to read the full text of a short article and ask them to come up with a catchy title.
↑This is an example. Now, I don’t like this title. It’s like, “Wouldn’t you like to read a book with that title?” and the sense of naming that seems to look right through you. It is a pattern I see in a lot of web articles these days, and in fact I tend to click on it. However, it is not my intention, because I do not want that kind of art style myself. Especially since “Oca Deus” is a category that I started with the intention of writing more introspective articles, this kind of title is not what I intended.
However, I don’t have any consistent beliefs about this site myself, so I accept it thinking, “I don’t like it, but you did a good job of coming up with this sentence. That in itself is interesting, so let’s use it.
(The above is the text that was proofread by ChatGPT and then fine-tuned for shaky notation by Okaden. The process of creating Okaden, proofreading by ChatGPT, and fine-tuning by Okaden is labor-intensive, not labor-saving. (Also, the splitting of sentences and copying and pasting is all done by hand.)
English website is now available.
English website is now available.
A link to switch languages is also located in the menu in the top right corner of the screen. However, it may not be visible when viewed from a smartphone.
The reason why we thought of doing this was because of an incident. That is, in early April, the number of hits on this site suddenly increased dramatically. I was alarmed and wondered, “What’s going on, are we being targeted by something? I was alarmed, but it was simply because the article had become a buzzword.
I am not sure why this article was buzzed about. I am not going to ego-search to find out. I don’t want to see criticism after searching this way and that. In any case, I got greedy because the number of hits was great and the ad revenue was accordingly high.
I have lost my desire for traffic in the last 10 years, but it has been a long time since I had the desire. I do not intend to write buzz worthy articles as it is not in my nature to do so. The most sincere answer would be to improve the frequency of publication and constant updating of the articles I’m currently working on, but that doesn’t excite me.
Then we turned to text editing with ChatGPT and multilingual support.
Multilingual support. We completely licked it.
Since it is impossible to manually translate every single word, we wanted to use machine translation from the beginning. However, we don’t want to just use a browser extension to translate a Japanese website and read it in a foreign language on the spot. We want to create a foreign language page that can be found by search engines based on search terms.
Then, of course, the number of pages will double with each additional language. Currently, awaremi-tai.com has just under 4,000 articles, so if all of them were translated into English, the number of articles would be 8,000. Even if we don’t translate all the old articles, because many of them will be out of compliance, we will have to translate more than 1,000 new pages.
I thought that with today’s technology, translation would be a breeze. In fact, there are tools that can do this and provide multilingual support. However, they cost quite a bit of money per month, so unless you are trying to monetize your site with international traffic, using a paid translation service is a waste of money. At the very least, this site can never use a paid translation service because the content is impossible to monetize with multilingual support.
Still, I thought I would try to do as much as I could for free and only translate the most recent articles, so I started working on it. I immediately regretted it. I thought, “I’ve gotten myself into a lot of trouble.
Before we can start translating an article, we have to translate the Japanese that appears all over the page. There are so many places where machine translation is not possible, such as the names of corners.
Then there is the translation of the article, which is the main part of the project. The two steps of translating from Japanese to English and proofreading the translated English are done by an outside service, so there are a lot of copy and paste bucket relays. In addition, since we are using a free service, we have to go through the trouble of breaking the article into smaller parts because of the strict character limit.
I thought about signing up for a paid service to reduce some of the translation work, but it still seems impossible to batch replace sentences.
For example, the Japanese word “Take” can be translated as “My son, Take”, or it can be translated as “Takeh”, “Takeshi”, or “Bamboo”. In other words, the results of the translations are different. One of the characters, “Ishi,” is sometimes translated as “Ishi” or “Isi,” and sometimes as “she” or “he” in the third person. It is true that the fact that “Ishi” is a woman is not learned by the translation tool, so it is natural that the translation results are fuzzy. However, the meaning is unclear to the English reader.
That is still fine, but an article about a visit to a restaurant is full of proper nouns, including the name of the dish. Sometimes it is not possible to figure out how to translate the names of foreign dishes into English at all, even if they are Japanese dishes. For example, “Chacharugun,” a drink served at a Mongolian restaurant, is a Mongolian word translated into Japanese katakana, but this katakana cannot be translated directly into English as “Chacharugun.
I had originally intended to translate the mountain and hot spring travelogues in “Drunken Travelogue” and the food series in “Declaration of Stomach Supremacy” into English. Other than that, I did not get around to it, and I do not think there is a demand for it in English. However, I do want to make all of the current mass produced articles at least available in English as soon as they are published, and as a result, many of the “Oka-Deus” articles are in English.
In truth, “Oka-Deus” is a corner I do not want to put in English. What I unknowingly do to my children may look like “child abuse” in other countries. Still, I can’t help it, my latest updates are all articles about Oka-Deus.
The real content of this site is the “AWAREMI-TAI Activity Record”. The other sections are just additions to this section. However, AWAREMI-TAI was active in the 90’s and 00’s, and there may be many problems if the texts written in those days are brought into the limelight again in 2023 and translated into English. Therefore, translation is not being considered at this time. There are many radical things written in jest, and because there are many conversational sentences, it would be difficult to translate.
Extreme expressions may still be acceptable if they are words like “die,” but the possibility cannot be ruled out that they may be unintentionally mixed with discriminatory remarks against minorities and other groups. There are many things that used to be considered acceptable jokes that are now considered NG. Even rewriting Japanese in Japanese is difficult, but translating it into English can be even more dangerous.
At first, I thought it would be great if I could handle “Chinese (Traditional),” “Chinese (Simplified),” and “Korean,” but after 15 minutes of creating the English home page, I realized, “Oh, this is impossible. I could tell something was wrong with the translation even with my English level, but when it came to Chinese and Korean, I couldn’t tell the difference. I had to give up.
I will only translate past writings into English in moderation. As I write this now, I think I should probably stop writing “Declaration of Stomach Supremacy” because it is probably tedious as well. This is because the article contains a lot of content, such as conquering all menus and walking around eating in restaurants, and a lot of proper nouns are used.
What about future articles? I have already started to make changes in the Japanese writing phase. Instead of writing without thinking, as I have done in the past, I am now using a writing style that is conscious of the fact that it will be translated. Specifically, I try to reduce the number of non-subjective sentences as much as possible and refrain from using colloquialisms and Internet slang.
For example, in “Oka-Deus,” the term “my son,Take” used to appear many times in a single article, but in recent articles, the term “My son,Take” has appeared only once in the first article, and after that, the expression “he” has been used. This is to avoid complicating the translation.
Writing with an English translation in mind completely diminishes my unique artistic style and momentum, but makes it easier to translate. What am I going to do?
Perhaps in the not too distant future I will stop translating into English and running an English website because it becomes too much of a hassle. Until then, I will try to go on like this for a while. I wonder how long that will last.