Dex-chan lover
- Joined
- Apr 10, 2023
- Messages
- 1,640
TL;DR: You're best off using it as a second opinion for a difficult line.
I was overwhelmed with curiosity with regards to a novel, so I decided to tackle it with pure MTL. At first, I used Google translate to tackle the pages in chunks, but Google is exceptionally bad at context, which basically makes it useless to get whole sentences out of natural dialogue in Japanese; the thing is physically incapable of keeping track of people, so the subjects and objects are always off somehow. Between that, and the character limit, I decided to spring for something I knew would be better: DeepL. So I went over and paid for a month. Immediately, I could put entire pages in there and (I thought), get a rough understanding of what's going on.
I could... except there was a problem: DeepL either works, or it doesn't try at all. Example:
It frequently gets caught in these weird loops, where it'll ignore the original text altogether and just spout some variation of the same (entirely unrelated) lines. At least once a paragraph, I get something like this. If it's not this, it's some variation of "The most important thing is...", or "The actual 'I don't like the way that...' ", among others. You can't fix this. There's no way to highlight the text as wrong, you can't ban certain translations and demand the raw if they come up, and you can't even find out what moon logic led to the imaginary translation, because it doesn't actually highlight what part of the sentence it thinks corresponds to what else. (For the record, this line is basically urging someone else hurry up before lunchtime hits). I'm forced to turn back to Google Translate to get the line if I don't want to spend an extra three minutes analyzing the clearly-ignored chicken scratch. At which point, I start to wonder why I paid money for this. The other issues are context (But less so, if there are hints present in the exact same sentence), and stuttering.
You can't Ctrl+F to find a specific string of text, and while you can also just have the text read out to you, you have to hear it all from the start: No picking out specific lines to read. Besides, I didn't pay money to get glorified story time.
But then at the same time, it has a clear grasp of the rules of Japanese, and can turn particularly particle, and adjective-heavy sentences into relatively natural, readable text that, upon close inspection, actually does jive with the story. When it actually tries to translate most other text, while imperfect, it's just the thing you need to speed through literal novels worth of information in a reasonable amount of time. I likely never noticed the problem because with the exact volume and length of the text I submit, the program rarely needs a GT's help.
Honestly, the only reason I've even kept it is because of the sheer volume of chicken scratch I have to get through. If It wasn't a literal series of novels, I would have gotten the refund ages ago and just stuck to the free version. So I guess I'm cancelling next month, after paying them properly for this one.
I was overwhelmed with curiosity with regards to a novel, so I decided to tackle it with pure MTL. At first, I used Google translate to tackle the pages in chunks, but Google is exceptionally bad at context, which basically makes it useless to get whole sentences out of natural dialogue in Japanese; the thing is physically incapable of keeping track of people, so the subjects and objects are always off somehow. Between that, and the character limit, I decided to spring for something I knew would be better: DeepL. So I went over and paid for a month. Immediately, I could put entire pages in there and (I thought), get a rough understanding of what's going on.
I could... except there was a problem: DeepL either works, or it doesn't try at all. Example:
It frequently gets caught in these weird loops, where it'll ignore the original text altogether and just spout some variation of the same (entirely unrelated) lines. At least once a paragraph, I get something like this. If it's not this, it's some variation of "The most important thing is...", or "The actual 'I don't like the way that...' ", among others. You can't fix this. There's no way to highlight the text as wrong, you can't ban certain translations and demand the raw if they come up, and you can't even find out what moon logic led to the imaginary translation, because it doesn't actually highlight what part of the sentence it thinks corresponds to what else. (For the record, this line is basically urging someone else hurry up before lunchtime hits). I'm forced to turn back to Google Translate to get the line if I don't want to spend an extra three minutes analyzing the clearly-ignored chicken scratch. At which point, I start to wonder why I paid money for this. The other issues are context (But less so, if there are hints present in the exact same sentence), and stuttering.
You can't Ctrl+F to find a specific string of text, and while you can also just have the text read out to you, you have to hear it all from the start: No picking out specific lines to read. Besides, I didn't pay money to get glorified story time.
But then at the same time, it has a clear grasp of the rules of Japanese, and can turn particularly particle, and adjective-heavy sentences into relatively natural, readable text that, upon close inspection, actually does jive with the story. When it actually tries to translate most other text, while imperfect, it's just the thing you need to speed through literal novels worth of information in a reasonable amount of time. I likely never noticed the problem because with the exact volume and length of the text I submit, the program rarely needs a GT's help.
Honestly, the only reason I've even kept it is because of the sheer volume of chicken scratch I have to get through. If It wasn't a literal series of novels, I would have gotten the refund ages ago and just stuck to the free version. So I guess I'm cancelling next month, after paying them properly for this one.
