How can my mistakes be corrected on Langua?

In Chat Settings, choose your preferred style from the 'Correction Mode' dropdown. There are five options, so you can explore and see what works best for you:

- Off: no interruptions during the chat, best for keeping the conversation flowing. You can still get feedback afterwards (see further below).

- Text corrections: see corrections as words crossed out and bolded beneath your replies. Turn this on, or tap the pencil icon below any reply to check it on demand.

- Implicitly correct me when possible: the AI naturally works the correct form into its own reply (e.g. echoing back what you meant to say), so you pick it up without breaking the conversation.

- Explicitly correct me: the AI gently points out your errors as part of its response, then carries on.

- Ask me to repeat: the AI pauses and asks you to say the corrected phrase back before continuing.

These auto-correction modes are switched off in roleplays (where auto-correct would mess up the roleplay), as well as in the guided course, grammar exercise and the custom prompt sections, where the correction behaviour would conflict with the exercises.

When you end the chat and have replied at least 6 times, you'll also see an option to generate a feedback report. These reports are popular as they provide detailed feedback, suggestions and even cultural tips.

Take corrections with a pinch of salt

Sometimes, you may see that the AI gives different suggestions depending on the method you use. This is because each method approaches it from a different angle. Also, AI models are not perfect and may make the following errors occasionally:

a) correct you when they shouldn't (for example, it may correct the capitalisation of a word when it doesn't remember that you're speaking, not writing), or it may suggest a more formal version when yours was fine. When you view corrections via the pencil button, you will note a 'check & explain' link. Tap that and the AI will explain why it has made the corrections, as well as check its answer. Occasionally, it may spot that it had made an error.

b) miss a correction (in our testing it does spot the errors around 95% of the time)

c) make typos e.g. a duplicate word is visible after it crosses something out.

We expect these imperfections to decrease over time as the models continue improving. We're using the world's best available models and are constantly testing and optimising.

If you doubt that a correction is right, one option is to actually ask the AI in your next reply. You can say it partly in your native language if needed e.g. 'I was told that I should use X instead of X but I'm not sure that's correct.'

Note that if the correction in question was provided in writing below your response (via the pencil button), rather than verbally, then the AI won't know about that correction, so you will have to give it enough information. This is also why a verbal correction may not always match the written correction shown via the pencil button.

Hope this information helps!