It has been a year since I started using ChatGPT. I mainly use it for programming and language translation, and the translation tasks, in particular, have been incredibly efficient.
Recently, I was able to make my Voicepaper app support 20 languages at once. ListTimer and Zeny can now support 12 languages. To be precise, I’ve had the help of an exceptionally talented person to multilingualize screenshots and text using ChatGPT, but the translation part is entrusted to AI, which has significantly reduced both speed and cost compared to before.
What’s different from the Google Translate era is the translation accuracy. When translating from English to Japanese, the awkwardness is significantly lower, and the same goes for the reverse. It has improved from a barely usable level to a practically usable one, so I started using it extensively.
I also compared it with DeepL, but personally, I find ChatGPT to be much better in terms of accuracy and usability. The best part is that you can copy and paste data files as they are and ask it to translate only the necessary parts while maintaining the format.
“RepeatThePlaylist” = “現在のプレイリストをリピート”;
⇩
“RepeatThePlaylist” = “Repeat The Current Playlist”;
The Cost of Multilingualizing Apps Has Dropped Significantly
Actually, a few years ago, I used a service called Gengo.com to multilingualize ListTimer into about seven languages at once. I would send the app’s language files with instructions on which parts to translate and which parts to leave as they were, and each file was translated at a cost per word.
This process was costly, had to wait several days for delivery, and updating all languages for minor updates within the app was extremely challenging. In the app world, even if you support eight languages, it’s often a matter of luck which language will drive downloads, so the cost and return did not match at all, and it was a failure.
However, the advent of generative AI solved this problem at once. It can handle it instantly and automatically determine the structure of the file for translation.
Moreover, translating buttons and messages for tool-type apps, unlike novels or movies, doesn’t have many critical scenes where the atmosphere would collapse significantly. Even if there’s a slight error, users can infer the meaning and use it. Therefore, except for a few important parts, it mostly suffices.
The End of Japan’s Galapagos Strategy?
Until now, Japanese web services have been viable due to the entry barrier of the Japanese language and the country’s considerable economic power. They could mimic high-level overseas services and create Japanese versions that could be commercially successful even at a moderate level.
However, from now on, developers from English-speaking and Chinese-speaking regions can easily support Japanese using AI. With low costs and high accuracy, the Japanese language barrier will gradually disappear.
Before rejoicing that it will become easier to sell in overseas markets while in Japan, there is a fear that most apps and services sold in Japanese stores will become foreign-made. While there are many services that are difficult to proceed without understanding unique Japanese customs, for simple apps, if the localization is decent, many cases will opt for those. In fact, the users of my household accounting app Zeny have become mostly overseas without me noticing.
Using AI for Blog Translation Can Lead to Ten Times the Readership
Translating blog articles using AI is even more straightforward than localizing apps. Previously, machine-translated blog articles were not at a readable level.
With the advent of ChatGPT, it translates to a level that can be published as is. Blog articles often get read randomly, so multilingualizing them becomes a great strategy.
In the past, I tried translating Japanese to English on platforms like overseas crowdsourcing sites, but it was costly and tedious. Now, using generative AI, I can translate and post them quickly.
For example, I wrote an article on a whim, translated it into English with ChatGPT, and posted it on HackerNews, which surprisingly went viral. It was entirely translated by generative AI.
While an article in Japanese might get about 100 views in a day, in English, it reached about 10,000 views. So, if you have a website or a blog, it might be a good idea to translate it into English.
Most cases will go unnoticed, but you never know what will hit, and the cost of buying that lottery ticket has drastically decreased.
*Notes
- I’m using ChatGPT4, and when there are limitations, I use a service called TypingMind to use it via API.
- Currently, the number of tokens for reading long texts is not an issue, but the output of translated texts stops halfway if the text is too long, which is challenging. Even with the latest Claude, it still stops. Google’s Gemini was unusable due to low accuracy. After trying various options, I settled on ChatGPT4.
*I've made Text-to-Speech, Money Tracker, and Timer apps. About Me.
