[en] Over-generated word forms

[cacfd3a]

Take as an example the Georgian term ამერიკის შეერთებული შტატები

Wiktionary only lists 'ამერიკის შეერთებული შტატების(ა)' (ameriḳis šeertebuli šṭaṭebis(a)) for the singular genitive, but wiktextract seems to generate these bizarre forms which are even contaminated with Latin characters:

ამერიკის შეერთებული შტატების [genitive, singular], 
ამერიკის შეერთებული შტატების [genitive, singular], 
ამერიკის შეერთებული შტატებისაs) [genitive, singular], 
ამერიკის შეერთებული შტატებისაsa) [genitive, singular],

It should only show ამერიკის შეერთებული შტატების and perhaps the archaic form ამერიკის შეერთებული შტატებსა.

Perhaps the script is trying out the combinations with and without the final letter ა (a), depending on euphonia?

[cacfd3a]

Also, it should recognize the postpositional inflected forms, instead it gives them tags for 'error-unrecognized-form'.

[kristian-clausal]

The აsa) bits are most probably from trying to generate alternative forms based on the parentheses, which it does partially right by having the ა at the end... and then the s and a are added from the end of the šṭaṭebis(a) with the romanization, because there's too many parentheses and this specific case has not been kludged. So a typical "table does something weird that doesn't happen elsewhere", which needs to be kludged.

The unrecognized form tags are because the table is probably pretty new and we just need to add the data for the unique headers like -ზე (-ze, “on”), it just needs some work... EXCEPT now that I look a them, there is no singular linguistic term tag I can find for -ზე (-ze, “on”) in Georgian specifically. If you could provide some generally agreed "names" for these post-positions that use the usual kind of graecolatin linguistic terminology (like has anyone ever called -თან (-tan, “near”) an "adessive" ending or something like that?), that would be great.

We could also just ignore these if these are regular postpositions that are purely agglutinative. They're called adpositions/postpositions which seems to indicate they're not part of the word morphology, but they appear in this table which seems to indicate they're stickier than adpositions normally... The postposition doesn't seem to make any changes inside the preceding word or noun-phrase.

[kristian-clausal]

I just spent too long trying to figure out where exactly the parser inserts parentheses around the romanization in examples like 'ამერიკის შეერთებული შტატები (ameriḳis šeertebuli šṭaṭebi)'. Couldn't find it, was going mad.

It turned out to be something in the original wiki HTML output that should be there:
<td><span class="Geor" lang="ka"><a href="[/w/index.php?title=...](view-source:https://en.wiktionary.org/w/index.php?title=...)" class="new" title="ამერიკის შეერთებული შტატებისებრ (page does not exist)">ამერიკის შეერთებული შტატებისებრ</a></span> <span class="mention-gloss-paren annotation-paren">(</span><span lang="ka-Latn" class="tr Latn">ameriḳis šeertebuli šṭaṭebisebr</span><span class="mention-gloss-paren annotation-paren">)</span></td>

I assume some CSS or javascript stuff toggles annotation-paren or mention-gloss-paren on and off depending on user settings or something similar... Couldn't find any of that at a quick glance, but I wasn't going to dig deeper.

[xxyzz]

There is a CSS rule to hide the "span" element.

This could be solved by selecting the "span" nodes that have "lang" attribute, but might be difficult for en code...

[kristian-clausal]

I've been working on this intermittently, and doing it "properly" was not working out so I did it as a terrible, horrible, stupid kludge that should be removed. I've added a special case for Georgian, so shouldn't mess up other languages at least, and I might have missed some other tables that are now broken because of this. If you could keep an eye out on stuff, that would be great.

[xxyzz]

How about we use a separate function to extract the "ka-decl-noun" template table? We could probably find the template node at here:

wiktextract/src/wiktextract/extractor/en/page.py

Line 2464 in 0d0556c

text = wxr.wtp.node_to_wikitext(node.children)

[kristian-clausal]

That's probably the best way to go... Especially if we could start from a blank slate.

However I found something that I completely forgot about: TableContext. It's a context object for EN inflectional tables, and I think I might have been the one to implement it... It's got TableContext.template_name, so instead of checking lang == Georgian we can check for ka-decl-noun, and we can do it in the same place we handle all the other tables.

[kristian-clausal]

I'll leave this open for future changes if we take a deeper look at this, and to give a chance for OP to check if everything went all right.

[cacfd3a]

Thank you for your tireless efforts! I can't really follow the discussions on the Wiktextract internals, but can help out with your question:

If you could provide some generally agreed "names" for these post-positions that use the usual kind of graecolatin linguistic terminology

In fact it is common for Georgian to use the suffixes as descriptions because the meanings overlap, so you still need a way to disambiguate the e.g. "-ვით" from "-ებრ". The names below are necessarily improvised.

I suppose these names would work as canonical names, borrowed mostly from the Finno-Ugric grammar (you are all Finnish speakers?):

Case Type	Georgian Form	Romanization	Meaning	Canonical English Name
Dative-case postpositions
	-ზე	-ze	"on"	Superessive
	-თან	-tan	"near"	Adessive
	-ში	-ši	"in"	Inessive
	-ვით	-vit	"like"	Semblative
Genitive-case postpositions
	-თვის	-tvis	"for"	Benefactive
	-ებრ	-ebr	"like"	Semblative
	-კენ	-ḳen	"towards"	Lative
	-გან	-gan	"from/of"	Ablative
	-ადმი	-admi	"in relation to"	Perlative
Instrumental-case postpositions
	-დან	-dan	"from/since"	Ablative
	-ურთ	-urt	"together with"	Comitative
Adverbial-case postpositions
	-მდე	-mde	"up to"	Terminative

[cacfd3a]

if these are regular postpositions that are purely agglutinative

I believe they are completely regular, yes, with the caveat that post-positional suffixes attach mostly to the oblique root but some attach to the nominative case. So it is still useful for a dictionary to show them (although again, this is completely regular).

For example:

• ქარივით (karivit) - "like wind": -ვით (-vit) attaches to the full nominative form ქარი (kari)
• ქარში (karshi) - "in the wind": -ში (-shi) attaches to the oblique stem ქარ- (kar-)

I also thought it might happen that phonotactic constraints might change the root (i.e. adding or eliding a vowel) but I struggle to come up with an example; even these phonemic rules are predictable. But I'm not an expert in Georgian: surely one who is, could come up with an exception to the rule.