Japanese input on Duolingo

I’m trying to use LipSurf for my Japanese lessons on Duolingo, and with version 4.12.1, it is almost working. However there seems to be a problem with conversion between kanji and kana.

Most of the time, I can see that the live transcript matches what I’m saying, and uses the correct kanji. But most kanji get replaced with hiragana along the way, and in the text input box I often end up with nothing but hiragana.

What’s worse, those hiragana are often incorrect readings of the kanji, and don’t match what I actually said. For example, in one answer I used the word もっている, which was correctly spelled 持っている in the live transcript. But in the input box I then got じっている, which, as far as I can tell, is not even a word (although じ is a possible reading of the kanji 持).

Does this happen to anyone else? Is there any way to fix this?

1 Like

this is an artifact with the way the speech-to-text works with Japanese. Will Duolingo accept answers in kanji? Or must it always be in hiragana?

In general, answers in kanji work just fine, and pretty much all Japanese text on Duolingo beyond the first few lessons uses kanji where appropriate.

I did encounter a few cases where Duolingo requires certain words to be spelled in kana, whereas my IME favors the kanji spelling, but this is pretty rare. Also, there are quite a lot of instances where Duolingo rejects synonyms, alternative grammar etc., so kanji vs kana seems like a relatively minor issue in comparison.

I can see why a phonetic spelling using hiragana might be preferable from LipSurf’s point of view, as that would usually be the “safest” choice. But that doesn’t help when the voice input has already been mangled by converting to kanji and back. And in those cases where I need to reach for the keyboard to tweak my answers, I’d much rather do so in a readable sentence, including kanji.

1 Like

Yeah this is more complicated than it seems because the speech-to-text engine always returns kanji. So LipSurf has to infer kana from there and ofc it can be ambiguous. I think the only good solution here would be to use the kanji directly from the speech-to-text engine on free response questions - that’s where you’re editing the answers, right?