Apr 4, 2021, 6:57 PM

Since people posted here about DeepL, I gave it a try.

My impression is that DeepL gives a adequate translation about 85% of the time, makes bad inferences 10% of the time, and the remaining 5% it just makes stuff up that has nothing at all to do with the original text.

Here's an example of it making stuff up, from Town Building Game web novel.

JP:

常に余裕がある表情を心掛けて、笑顔を絶やさぬように気を張っていたので表情筋が固まってしまった。

がさっ、という物音と視線を感じたので足下に向けると、ディスティニーが俺を真似して自分の頬をモミモミしている。

……くっ、ちょっとかわいいと思ってしまった。

DeepL:

I was always trying to keep a relaxed expression on my face, and I was so careful to keep smiling that my facial muscles froze.

This is a great way to make sure you are getting the most out of your time with your family and friends.

I thought it was a little cute. ......

Me:

I'd been trying so hard to smile like I had everything under control that the muscles on my face had hardened.
I felt something at my feet, and when I looked down, Destiny was imitating me, giving himself a cheek massage.
I thought that was a little cute.

I think my verdict is that, as with other MTL, use DeepL with caution and as a last resort. You'll need to know the source language anyway so you can fill in the gaps when the MTL makes mistakes, and if the MTL makes only small mistakes, you won't notice.