volatile memorandum

Home Blog

• 14 Dec, 2025 *

I’ve been having good conversations through language exchange Japanese (Native) and English (B1 or B2). It’s a nicer experience than AI chat or AI interviews.

Recently I’m interested about AI translated Japanese and noticed that are correct but not natural. I’m guessing Japanese learners might be affected by that.

But what is the difference, and where does it come from?

Disclaimer: I’m not a linguist and not well educated.

Too much correction of 助詞

AI (e.g. ChatGPT) uses 助詞 well but sometimes feels unnatural because too correct. Almost Japanese omit some 助詞 unintentionally because we don’t use the grammer checker tools even when writing e-mail. So I think AI should do the same in casual conversation.

You might be able to omit 助詞 (in casual chat)

「もし刀削麺1（が）好きなら、明後日（に）秋葉原（の）刀削麺屋（に）行かない？」 but if omitted the mid "の" turns unnatural.

Too much use of 終助詞

AI seems to use ～ね and ～よ too much like "今日はいい天気だね。" or "～だと思うよ。" These 終助詞 give a warmer and friendlier nuance but if heavily used I feel pretty weird. I sometimes use them as an accent or when I want to emphasize something.

Mixing styles such as 口語, 文語 and 敬語

You should avoid it unless know the nuance. But even I do it carelessly when writing technical documents.

I guess AI does not grasp these nuances perfectly yet. DeepL’s completion feature is not aware it so can create strange sentences.

This is my favorite 刀削麺屋 刀削麺荘 唐家 秋葉原本店↩