Spellchecker improvements

[MXC48]

Some things in this list are more or less important to add, they are just ideas.

• make sure that the keyboard no longer suggests words with a capital letter (except for proper nouns), nor corrects words by adding a capital letter inside a sentence.
• set up a black list for words that the keyboard will never learn, to avoid regularly misspelled words ending up in the system dictionary.
• ensure that the keyboard directly recognises all digits and numbers without underlining them as wrong, and that words followed or preceded by punctuation are not always marked as wrong either.
• ensure that verb variants with reflexive articles (particularly in French) are recognised by the keyboard, because at the moment not all of them are, and this can often lead to incorrect corrections (examples of reflexive articles: m'; t'; s').
• it would be good to allow the community to improve the downloadable dictionaries by allowing new words to be added or even with the addition of another downloadable dictionary adding lots of words that aren't in the oldest one (for French here because I see that for others they draw their sources from the wiktionary).
• for French here in particular because English has fewer difficulties as it has fewer conjugations, add the conjugations for many verbs in the keyboard dictionary or another.
• when there is text in the clipboard, automatically suggest it in the keyboard toolbar when the clipboard is opened.

I hope you'll agree with me on the usefulness of these things 😁 for me it would drastically improve my comfort of use on a daily basis! I'm well aware that there are many other things to do before adding all this and see this more as a list of ideas that can evolve over time!
(PS maybe there's some duplication with other things I've added)

[Uranusek]

These underlined numbers are sometimes annoying. Single letters are also underlined, only the English I is not. Once these numbers and letters are added to the personal dictionary, they are no longer underlined, but that's not the solution.

There's one more thing that irritates me about spellchecker:
If I start typing a word with a capital letter, accidentally tap Shift and add a capital letter in the middle of the word, I only get suggestions with a lowercase letter. To get a suggestion with a capital letter I have to delete the whole word and start typing it again. It's a bit annoying when I start writing a sentence.

[realpixelcode]

• set up a black list for words that the keyboard will never learn, to avoid regularly misspelled words ending up in the system dictionary.

For an explicit list of incorrect words that could be included in such a blacklist, here are a few resources that might be helpful:

Duden

The Duden is a well-known German dictionary. On their website, they provide a list of frequent spelling mistakes. (Note that the words listed there are the misspelt versions.) As far as I can tell, all of the misspellings there are absolute/clear mistakes.

korrekturen.de

This is a website of an editorial office that also maintains a list of German words frequently misspelt (“popular errors”). Knowledge of the German language or use of a website translator may be needed to navigate the site and distinguish between clear/absolute errors and “rather wrong”/non-absolute ones. (One reason such non-absolute mistakes exist at all is that there was a government-mandated 🤡 spelling reform twenty years ago.)

Grammle

It's a project (of mine 😉) that lists frequent German spelling, grammar, style and other linguistic mistakes. In the list of orthography mistakes, the colour red is used for absolute mistakes (unquestionably/always wrong, according to sources), whereas yellow marks mistakes that sources consider rather wrong (in contrast to “clearly wrong”) or not recommended.

Note that in the other categories (especially grammar/Grammatik, meaning/Bedeutung and style/Sprachstil), there may also be some candidates for being included in a blacklist. (Consolidated list)

The full list of mistakes is also available as a YAML file. Look for items where the kategorien property includes orthographie or where the ID starts with o. The arten property of each item tells you whether it's an absolute mistake (fehler), a non-absolute one (warnung) or just not recommended (info).

[Staubsaugerbeutel]

One of the most annoying autocorrect behaviours (especially when set to aggressive) is that it contains so many acronyms that aren't proper words and accepts misspells and capitalises them if there's some acronym that happens to be the word you misspelled instead of replacing it with a correct word. At this point I manually removed all commonly appearing acronyms but it would be nice to be able to just remove all of them.

[realpixelcode]

I noticed that the spell checking has begun automatically capitalising the German personal pronoun “ich” of the first person singular even when it occurs in the middle of a sentence (i.e. where it has to be in lowercase) 🤔

[MXC48]

@Helium314 are these suggestions possible? Through dictionaries, for example?

[Helium314]

I converted this to a discussion, since it does not work as an issue. Especially because it's a bunch of loosely related things.

[Helium314]

• make sure that the keyboard no longer suggests words with a capital letter (except for proper nouns), nor corrects words by adding a capital letter inside a sentence.

When does the keyboard add unwanted capital letters? Is it words that are in a used dicitionary?

[MXC48]

I've already noticed the problem when I was using the stable main.fr dictionary, and the problem persists with the experimental version.
At the moment it corrects "ça" to "Ça" when it shouldn't, it can also do it for :
"je->Je"
"moi->Moi"
"va->VA"
(I'll try to be careful and I'll add to the list as I go along to give a good idea).
I don't know if the problem exists for other languages

[Helium314]

I'll have a look whether the words are in the dictionary.
For the experimental ones I think there is an issue that capitalized versions of words are sometimes added, at least I remember something like that for the English one.

If not, next thing is looking at raw suggestions provided by native code.

[Helium314]

• set up a black list for words that the keyboard will never learn, to avoid regularly misspelled words ending up in the system dictionary.

I can't seriously set up such blacklists

• people may even want to use a word that is a regular spelling mistake
• it's a lot of work, and requires knowledge of the languages. I simply don't have the resources

[MXC48]

I understand the complexity of the task, so as a first step it might be a good idea to allow users to make a blacklist of words that would never be corrected and never recorded in the user dictionary.

[Helium314]

• ensure that the keyboard directly recognises all digits and numbers without underlining them as wrong, and that words followed or preceded by punctuation are not always marked as wrong either.

iirc there already is an issue for that, and I think behavior there was even dependent on Android version (things get worse at 10 or 11, I can't test as I'm still on 9)

[MXC48]

Is there any way I can make logs so that I can send them?
I'm running LineageOS 20 android 13 unofficial, redmi note 11

[Helium314]

I don't think the spell checker has much logging, even in debug mode.
The best way would be building your own version with a lot of logging in places that could be related to your issues.

[Helium314]

• ensure that verb variants with reflexive articles (particularly in French) are recognised by the keyboard, because at the moment not all of them are, and this can often lead to incorrect corrections (examples of reflexive articles: m'; t'; s').

You can add them to the dictionary. Either to you personal dictionary, or to a "proper" one in the dictionaries repository

[MXC48]

Where could I add these words? Despite my best efforts, I can't get to grips with python scripts 😅 if there's a list of words for French, I'd be happy to add them...

[Helium314]

There is no need for python scripts.
Best add the words to the main_fr wordlist (just the normal one). If you want to test it, you can compile the dict file with thedicttool linked in the dict repo readme.

[Helium314]

• when there is text in the clipboard, automatically suggest it in the keyboard toolbar when the clipboard is opened.

That should go into an issue (one for only this topic, see the text in the issue template: If you have more than one feature request, open multiple issues (one for each) unless the requests are closely related.)