I’ve noticed DeepL has some troubles while trying to translate Tategaki (vertical Japanese), as it follows in the screenshots.
This is the basis picture to translate.
This is the mistakenly translated Tategaki sentence.
Beside not translating the whole sentence, it displays incorrect kanji, such as 細 (/hoso/ = thin) over 話 (/hana/ = talk to), including 目 (/me/) that's simply nonexistent in the picture, and changing from 満 (/man/ = full) to 清 (/kiyoshi/ = clear). Seems like the API cannot recognise characters based on the typeface or something. Maybe not really trained enough for it.
Just like Arabic and Hebrew, the Tategaki logic is right-to-left, but the DeepL’s API treated it as if it was in the common order (left-to-right).
Estimated phrase:
「満月になったら、君について月とたくさん話すよ。」
Estimated translation:
“When it’s full moon, I talk to the moon a lot about you.”
I’ve noticed DeepL has some troubles while trying to translate Tategaki (vertical Japanese), as it follows in the screenshots.
This is the basis picture to translate.
This is the mistakenly translated Tategaki sentence.
Beside not translating the whole sentence, it displays incorrect kanji, such as 細 (/hoso/ = thin) over 話 (/hana/ = talk to), including 目 (/me/) that's simply nonexistent in the picture, and changing from 満 (/man/ = full) to 清 (/kiyoshi/ = clear). Seems like the API cannot recognise characters based on the typeface or something. Maybe not really trained enough for it.
Just like Arabic and Hebrew, the Tategaki logic is right-to-left, but the DeepL’s API treated it as if it was in the common order (left-to-right).
Estimated phrase:
「満月になったら、君について月とたくさん話すよ。」
Estimated translation:
“When it’s full moon, I talk to the moon a lot about you.”