#Any resources for grammar checking?
8 messages · Page 1 of 1 (latest)
<@&1248039856404431062> Any resources for grammar checking?
I've been wondering a similar question for myself, so I (just now) typed 文法チェック into google to see what I hit. There are hits but, hmm, the first one?
I gave it a bad conversion:
夢が敵う時は必ず来る should mean "the time when dreams come true is sure to come" but it's the wrong spelling of かなう (should be 叶う). I make this kind of error more often than I would like. Unfortunately it wasn't detected.
Next, a mistake that L2s often make and native speakers almost never do: 美しいだね - this is very cut-and-dry, no judgement needed. Also not detected. https://www.kiji-check.com/ I'm not impressed.
https://wordvice.ai/jp/tools/japanese-grammar-checker This one flags both errors, but also wants to upgrade the politeness of 来る to 来ます - useful for writing a business e-mail. Changing "document type" to "creative" stops it from doing that. Free registration required, not sure about usage limits, freemium model.
As a language-learning tool the best thing they can do is point your attention at things you already know how to fix. The guide/roadmap talks about competence and performance in stage 3. If you don't have competence you're not really able to second guess these (but you can ask a tutor or exchange partner). If it's a performance error they might be a lot more helpful. Before stage 3, when you don't have much grammar, they're not good enough to tutor you.
In Stage 2 it's a little too early to worry about writing correctly. To avoid bad habits spend a lot more time on input than output. Assume that everything you produce is questionable, but also that's kind of liberating: you probably can't fix it yet, so don't worry too much about trying.
I use this occasionally when I need to send text messages. It doesn't catch every error. For example, it doesn't catch Glass's first error. But it does catch the second.
Like Glass said, when you have more input, it will be easier for you produce correct output, and to tell when you're making a mistake, or why something is a mistake.
ChatGPT
Should be prompted in Japanese, hard to tell when it's overconfident. This might be a task that it can "zero-shot" (ugh, I hate that jargon) with better performance than you'd assume, but models really should be fine tuned for it. Biggest problem is that it makes mistakes with apparent confidence; here's a perfect example:
文「美しいだね」には少し文法的な修正が必要です。
正しい形は:
「美しいね」 または 「美しいですね」 です。
修正ポイント:
- 「だね」ではなく、「だ」は「です」や「だよ」に置き換える必要がありますが、形容詞「美しい」にそのまま「だ」を付けることは通常しません。形容詞「美しい」はそのままで「ね」や「ですね」で締めくくるのが自然です。
- もし、相手に対して敬意を込めたい場合は、「ですね」とするのがより丁寧です。
例えば:
- 「美しいね」(親しい間柄で使う)
- 「美しいですね」(少し丁寧な言い方)
どちらも自然な表現です。
||美しいだよ is not a good correlation, it's the same problem as 美しいだね||