[en] Various ancient Greek inflection errors

[zfletch]

This is a great project! I'm currently working on using the ancient Greek inflection tables to build a word-parsing tool, but I've found a few minor issues.

1. Definite article inconsistency

In some entries, the definite article is included in the inflected forms, but in some entries it is missing. This is the case even when the inflection tables in Wiktionary for both entries include definite articles. Definite article inclusion also differs for forms within a single entry. For example:

ὁδός

{
  "form": "τῇ ὁδῷ",
  "roman": "tēî hodōî",
  "source": "declension",
  "tags": [
    "dative",
    "singular"
  ]
}

{
  "form": "ὁδῶν",
  "roman": "hodôn",
  "source": "declension",
  "tags": [
    "genitive",
    "plural"
  ]
}

ἰχθυοκένταυρος

{
  "form": "ἰχθῠοκενταύρῳ",
  "source": "declension",
  "tags": [
    "dative",
    "singular"
  ]
}

2. There is no way to find verb tense

Even though the tense of a verb form is included in the Wiktionary page, there is no way to find that information in the extracted JSON. In Latin, the tense is included in the "tags" list for each inflected form. However, this may be because the Latin tables are formatted differently. Still, for ancient Greek nouns, the information at this level (the title of the table) can be extracted from the element with "tags": "table-tags". I'm not sure the best way to represent this information in the JSON, but I think it should be available.

ἐρωτάω (ancient Greek verb)

The tense, present, is nowhere in the JSON.

{
  "form": "no-table-tags",
  "source": "inflection",
  "tags": [
    "table-tags"
  ]
},
{
  "form": "grc-conj",
  "source": "inflection",
  "tags": [
    "inflection-template"
  ]
},
{
  "form": "ἐρωτᾰ́ω",
  "source": "inflection",
  "tags": [
    "active",
    "first-person",
    "indicative",
    "singular"
  ]
},

ὁδός (ancient Greek noun)

Although there is no tense for nouns, the information at the table title level (e.g. Attic) is in the JSON.

{
  "form": "Attic declension-2",
  "source": "declension",
  "tags": [
    "table-tags"
  ]
}

praerogo (Latin verb)

The tense for Latin inflected forms (e.g. present) is included in the tags.

{
  "form": "praerogāmus",
  "source": "conjugation",
  "tags": [
    "active",
    "first-person",
    "indicative",
    "plural",
    "present"
  ]
}

3. Incorrectly marked plural forms

Likely due to the structure of the inflection tables, some participial and infinitival forms have incorrect number. The infinitives should have no number (i.e. they should not say "singular", "plural", or "dual"), but some say "plural". The participles should all be "singular", but some are "plural" and some are lacking number entirely.

ἐρωτάω

{
  "form": "ἐρωτᾰ́ειν",
  "source": "inflection",
  "tags": [
    "active",
    "infinitive"
  ]
},
{
  "form": "ἐρωτᾰ́εσθαι",
  "source": "inflection",
  "tags": [
    "infinitive",
    "middle",
    "passive",
    "plural"
  ]
},
{
  "form": "ἐρωτᾰ́ων",
  "source": "inflection",
  "tags": [
    "active",
    "masculine",
    "participle"
  ]
},
{
  "form": "ἐρωτᾰόμενος",
  "source": "inflection",
  "tags": [
    "masculine",
    "middle",
    "participle",
    "passive",
    "plural"
  ]
},

4. (Possibly not fixable) Grammatical gender is difficult to find

This may be more of a feature request/complaint than an error, so feel free to ignore it. But it is difficult to find the grammatical gender of a word from the JSON. In the entry for ὁδός, for example, there is nowhere "feminine" appears. The only way to get the gender is to look at the "categories" and find something like "Ancient Greek feminine nouns" or parse the "head_templates" or "inflection_templates" and guess that if one of the arguments is "f", then the word is feminine.

[kristian-clausal]

Going through these issues slowly. For the articles, (sigh) it's that the tables on Wiktionary's side just do not stop changing. We have removed the articles from tables, but then the formatting changes or newer tables have some forms that we haven't taken care of, etc. etc. and suddenly it's broken. Thank you for bringing these changes to our attention, we will have to rely on people to keep an eye out for this going forward.

Please keep in mind we do not use 'AI' for this -- everything is hand-crafted, everything is done with manually curated lists of exceptions and hordes of kludges in the code that go "if we're looking at some Greek language table, remove all ^τοῦ, τῆς from the start of forms.

[kristian-clausal]

Going over these again, but this one...

3. Incorrectly marked plural forms

is really annoying to handle. If you look at how the table is structured, we just have a new section that starts and the two headers "active" and "middle/passive", but nothing else to indicate a new section has started that needs to reset headers. We do have tools to fix this, a long list of special rules (in a dict with its own control flow structures) that I could apply here... Except I can't 'attach' it to anything here. I could specify "for the template 'grc-decl', when you encounter the header 'active', reset headers..." except that messes things up because 'active' and 'middle/passive' are already headers in the earlier section, and doing that there messes things by removing all the headers before them.

[zfletch]

Thank you for looking into this! Sorry for creating such a long issue, but I didn't want it to look like I was spamming the repository by creating four separate issues.

Going over these again, but this one...

Incorrectly marked plural forms

is really annoying to handle.

It seems like it would be especially challenging for the participial forms. Infinitives don't have number, so if you know it's an infinitive, you know that those fields don't make sense. But participles do have number, and nowhere in the table is the number of those forms specified. The visitor to Wiktionary just needs to know that the dictionary form for a participle is the nominative, masculine, singular. (Edit: it looks like gender is given, but the reader still needs to know those are nominative singular forms.)

I don't know how the code works, but is there any way to look at the leftmost column to see if it's marked as infinitive or participle and "attach" the logic to that?

[kristian-clausal]

We would rather not add any logic that makes up data, because that's fraught with problems: if it's not in the Wiktionary data, we would rather not add it, even if it "should" be there. Especially in this part of code, it's too damn messy to add more language-specific should-be-kludges.

The header reset can't use the "infinitive" as its anchor point because it would reset all headers coming from above, including the previous line with "active" and "medial/passive". I'm trying to add a thing where I could check the column index in the row and use the "active" on that previous row to reset what is above it. The English extractor table parsing code is big and messy and really should have a good, long look at it (with an axe and flamethrower), but it is what it is because the English dictionary tables are genuinely a world of pain.

[xxyzz]

Maybe we could implement extractor function for each table template separately, some templates could use the same function if their table layouts are similar. If we have covered enough templates we can delete the old code altogether.

I could add a function for a table template to try to relieve this maintenance plain a bit but I'd like to hear your opinion and also don't want conflict with your unpushed changes.

[kristian-clausal]

I've just wasted a lot of time (well, I did other stuff too when I got frustrated) and couldn't figure out why my changes to inflectiondata conditional stuff wasn't working... and I had accidentally used grc-decl (from earlier) instead of grc-conj... In any case, after fixing that it worked fine, you can now check if a certain trigger phrase in a cell is found in the Nth column and do things based on that information. For rows, that's possible (I think) but it would need the addition of some parameter sprawl to keep track of the row number because we don't do that currently. Time to write some tests.

[kristian-clausal]

Sub-issues 1-3 seem to be okish now, might be missing some tags or table headers but those can be easily added in the future. Number 4 is a bug, because the head does have a f in it, but it might be ignored because it's between two parenthesized qualifiers... The forms wouldn't get the feminine tag, but the senses would.

EDIT: It's fine, even preferable to break up clearly separate issues into their own threads. If we used metrics at Clausal, that would definitely be one of them lol but it's also a bit cleaner when closing issues.

[kristian-clausal]

Let's never mention the issue with the bullet character in Greek heads again. 🤐

Most of these should be, if not completely fixed at least better. I'll leave the thread open and you can check in to tell if there's still issues with these specific things. If there's something else, feel free to make a new thread.

[zfletch]

Thanks! I've been working with the most recent JSONL and issues 2 (verb tense) and 3 (incorrectly marked plural forms) look like they've been fixed.

For issue 1 (definite article), I think there are still some problems, unless that change somehow hasn't been included in the current JSONL downloads. The λόγος JSON, for example, includes the definite article everywhere except for dual forms:

[kristian-clausal]

These just need to be added to the list, manually. If you find any more, post them here.