Casing in grammar checker

[flammie]

Currently the grammar-checker fails to realise the case in following:

$ echo an Uachtarán | tools/grammarcheckers/modes/glegram.mode
"<an>"
	"an" Art Sg Def <W:0.0> ID:1
:
"<Uachtarán>"		Uachtarán	→  T-uachtarán	→  TUachtarán
	"uachtarán" Noun Masc Com Sg <W:0.0> &msyn-noun-defart ID:2 R:$2:1
msyn-noun-defart
	"uachtarán" Noun Masc Com Sg <W:0.0> DefArt SUGGEST ID:2 R:$2:1
uachtarán+Noun+Masc+Com+Sg+DefArt	t-uachtarán,tUachtarán
:\n

$ echo an Uachtarán | divvun-runtime run -p tools/grammarcheckers/bundle.drb

[
  {
    "form": "Uachtarán",
    "beg": 3,
    "end": 13,
    "err": "msyn-noun-defart",
    "msg": [
      "msyn-noun-defart",
      "msyn-noun-defart"
    ],
    "rep": [
      "T-uachtarán",
      "TUachtarán"
    ]
  }
]

The expected result would be "tUachtarán", I think it needs some work on cg rules / lexc and some on libdivvun / divvunruntime. Is there existing codes that would work for this @unhammer ?

[unhammer]

It should work if you add a tag <fixedcase>.

You can quickly verify if you just add a rule
"<Uachtarán>" ADD (<fixedcase>) TARGET (Noun);

---

The test case has an example of fixedcase with SUGGESTWF like this:

"<ULBMILL>"
	"ulbmil" N Sem/Semcon Sg Nom <W:15.8955> <WA:7.89551> <spelled> "ulbmil"S <fixedcase> SUGGESTWF &typo #8->8

giving the suggestion "ulbmil".

[flammie]

yeah that does it, thanks, now I can avoid poking into libdivvun and/or divvun-runtime code pointlessly, and leave it up to others to implement the cg part :-)

$ echo an Uachtarán | divvun-runtime run -p tools/grammarcheckers/bundle.drb
[
  {
    "form": "Uachtarán",
    "beg": 3,
    "end": 13,
    "err": "msyn-noun-defart",
    "msg": [
      "msyn-noun-defart",
      "msyn-noun-defart"
    ],
    "rep": [
      "t-uachtarán",
      "tUachtarán"
    ]
  }
]

$ echo an Uachtarán | tools/grammarcheckers/modes/glegram.mode
"<an>"
	"an" Art Sg Def <W:0.0> ID:1
:
"<Uachtarán>"		Uachtarán	→  t-uachtarán	→  tUachtarán
	"uachtarán" Noun Masc Com Sg <W:0.0> &msyn-noun-defart <fixedcase> ID:2 R:$2:1
msyn-noun-defart
	"uachtarán" Noun Masc Com Sg <W:0.0> DefArt SUGGEST <fixedcase> ID:2 R:$2:1
uachtarán+Noun+Masc+Com+Sg+DefArt	t-uachtarán,tUachtarán
:\n

[unhammer]

Great :-)

Maybe something like a final

ADD (<fixedcase>) TARGET (SUGGEST) IF (-1 someword); 
# if we're not the first word of the sentence, use the case of the generator output

would be a start; I don't know how it should work for the very first word.

[uidhonne]

The forms t-uachtarán and tUachtarán only occur after the definite article an/An so it can be the first word in a sentence. Elaine.

…

________________________________ From: Kevin Brubeck Unhammer ***@***.***> Sent: Wednesday 26 November 2025 13:31 To: giellalt/lang-gle ***@***.***> Cc: Subscribed ***@***.***> Subject: Re: [giellalt/lang-gle] Casing in grammar checker (Issue #4) [External Email] This email originated outside of Trinity College Dublin. Do not click links or open attachments unless you recognise the sender and know the content is safe. [https://avatars.githubusercontent.com/u/56868?s=20&v=4]unhammer left a comment (giellalt/lang-gle#4)<#4 (comment)> Great :-) Maybe something like a final ADD (<fixedcase>) TARGET (SUGGEST) IF (-1 someword); # if we're not the first word of the sentence, use the case of the generator output would be a start; I don't know how it should work for the very first word. — Reply to this email directly, view it on GitHub<#4 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AEB4ZKQWFZL7JDJ6QAHPADT36WTSXAVCNFSM6AAAAACNH7L3AWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKOBRGMZTQNZQHA>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[unhammer]

So if there's a typo in the first word of a sentence, we can just treat casing like typos in first words in English etc.? E.g. in

"Neihbour told police the driver of the car in question was playing a recorder"

lemma is "neighbour" and divvun-suggest by default will apply titlecase because the form has titlecase.

[seananocoistin]

Thanks for the work and suggestions.

Further examples of this include:

"Sheol an Pharlaimint a tuarascáil chuig an gCoimisiún." The lowercase g before the uppercase C is correct.

an tsráid, An tSráid Mhór, is another example.

asal, an t-asal, an tAsal.

uachtarán, an t-uachtarán, an tUachtarán.

The code for the rules should be able to distinguish between the need for a lowercase T and a hyphen before a lowercase initial letter (e.g., an t-uachtarán) and a lowercase T and an uppercase initial letter with no hyphen used when the second letter is in the uppercase (e.g., an tUachtarán).

[uidhonne]

Yes, i think that that would be ok to treat a typo in the first word the same as for English, as lowercase prefixes to a proper noun such as eclipsis and h/n/t are all caused by a previous word. Elaine. So if there's a typo in the first word of a sentence, we can just treat casing like typos in first words in English etc.? From: Kevin Brubeck Unhammer ***@***.***> Sent: Wednesday, November 26, 2025 3:06 PM To: giellalt/lang-gle ***@***.***> Cc: Elaine Ui Dhonnchadha ***@***.***>; Comment ***@***.***> Subject: Re: [giellalt/lang-gle] Casing in grammar checker (Issue #4) [External Email] This email originated outside of Trinity College Dublin. Do not click links or open attachments unless you recognise the sender and know the content is safe. [An íomhá bainte ag an seoltóir.]unhammer left a comment (giellalt/lang-gle#4)<#4 (comment)> So if there's a typo in the first word of a sentence, we can just treat casing like typos in first words in English etc.? E.g. in "Neihbour told police the driver of the car in question was playing a recorder" lemma is "neighbour" and divvun-suggest by default will apply titlecase because the form has titlecase. — Reply to this email directly, view it on GitHub<#4 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AEB4ZKRJGOJ7LHPSJMMPDXL36W6VZAVCNFSM6AAAAACNH7L3AWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKOBRG4ZTSOBSGQ>. You are receiving this because you commented.Message ID: ***@***.***>

[seananocoistin]

It works. Thank you. There is one remaining issue and that is that if a noun with an uppercase letter is analysed, the result is given with a lowercase letter, e.g., Áras an Uachtarán becomes Áras an uachtaráin. It is a small thing, but it would be good to have the same case as the initial input.

Another example is "chuig an Coimisiún" becomes "chuig an gcoimisiún".

[unhammer]

Another example is "chuig an Coimisiún" becomes "chuig an gcoimisiún".

Then I would guess we should not have <fixedcase> on that one. So the CG rule needs to be more nuanced, to be able to differentiate between those two situations. Is there anything in the analysis input you could use to differentiate when you want to apply titlecase vs not?

        "coimisiún" Noun Masc Com Sg DefArt &msyn-ecl-after-prep
msyn-ecl-after-prep
        "coimisiún" Noun Masc Com Sg  Ecl SUGGEST
coimisiún+Noun+Masc+Com+Sg+Ecl  gcoimisiún

vs

        "uachtarán" Noun Masc Com Sg &msyn-noun-defart
msyn-noun-defart
        "uachtarán" Noun Masc Com Sg DefArt SUGGEST
uachtarán+Noun+Masc+Com+Sg+DefArt       t-uachtarán       tUachtarán

E.g. do you maybe only want <fixedcase> on the DefArt suggestions or something like that?

(Or are you also currently getting "t-uachtarán" and you want "t-Uachtarán" or something?)

[snomos]

It works. Thank you. There is one remaining issue and that is that if a noun with an uppercase letter is analysed, the result is given with a lowercase letter, e.g., Áras an Uachtarán becomes Áras an uachtaráin. It is a small thing, but it would be good to have the same case as the initial input.

Another example is "chuig an Coimisiún" becomes "chuig an gcoimisiún".

Are you talking about the form in the morphological analysis? If so, the lemma in the analysis is always given using lexical case, irrespective of syntactic context. This is part of how the system works, and can not be changed. Correction suggestions should be given with a case appropriate for the context, though.

[unhammer]

I interpreted it as being about the form in the final suggestion, ie. what's shown to the user?

When tagging a reading with <fixedcase>, the suggestion mechanism will simply use the output of the morph generator unchanged. When I try with the newest giella-gle build (8033c87) I get

coimisiún+Noun+Masc+Com+Sg+Ecl  → suggest gcoimisiún or gCoimisiún

uachtarán+Noun+Masc+Com+Sg+DefArt   → suggest    t-uachtarán or tUachtarán

So the FST currently does that second-letter uppercasing of the DefArt. And that seems good – such complex titlecasing rules should be the responsibility of the FST (or some part of the language-specific pipeline), divvun-suggest shouldn't have language-specific capitalisation rules beyond basic titlecasing.

So one issue is to decide for what cases you do want divvun-suggest to do basic titlecasing instead.

Another is: Do you want to prioritise gCoimisiún over gcoimisiún in titlecase contexts?