Enable the configuration created in T297761 on Hindi-language wikis.
Description
Description
Details
Details
Related Changes in Gerrit:
| Subject | Author | Repo | Branch | Lines +/- | |
|---|---|---|---|---|---|
| search: enable Latin-to-Devanagari transliteration second-chance | DCausse | operations/mediawiki-config | master | +11 -0 |
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | None | T375215 [EPIC] Support "second-try" transliteration or wrong-keyboard searches (aka N.O.R.M.) | |||
| Resolved | Feature | TJones | T297761 Create a Latin-to-Devanagari transliteration second-chance search for Hindi wikis | ||
| Resolved | dcausse | T425018 Enable Latin-to-Devanagari Transliteration second-try search on Hindi Wikis |
Event Timeline
Comment Actions
Change #1283041 had a related patch set uploaded (by DCausse; author: DCausse):
[operations/mediawiki-config@master] search: enable Latin-to-Devanagari transliteration second-chance
Comment Actions
Change #1283041 merged by jenkins-bot:
[operations/mediawiki-config@master] search: enable Latin-to-Devanagari transliteration second-chance
Comment Actions
Mentioned in SAL (#wikimedia-operations) [2026-05-06T07:31:24Z] <awight@deploy1003> Started scap sync-world: Backport for [[gerrit:1283101|VE: Avoid counting all refs when listIndex is undefined (T425433)]], [[gerrit:rCICF12830376bd04|search: fix alt. completion indices to test keyword tokenizer (T420427)]], [[gerrit:1283041|search: enable Latin-to-Devanagari transliteration second-chance (T425018)]]
Comment Actions
Mentioned in SAL (#wikimedia-operations) [2026-05-06T07:33:20Z] <awight@deploy1003> wmde-fisch, awight, dcausse: Backport for [[gerrit:1283101|VE: Avoid counting all refs when listIndex is undefined (T425433)]], [[gerrit:rCICF12830376bd04|search: fix alt. completion indices to test keyword tokenizer (T420427)]], [[gerrit:1283041|search: enable Latin-to-Devanagari transliteration second-chance (T425018)]] synced to the testservers (see https://wikitech.wikimedia.org/wiki/Mwdebug). Changes can
Comment Actions
Mentioned in SAL (#wikimedia-operations) [2026-05-06T07:40:23Z] <awight@deploy1003> Finished scap sync-world: Backport for [[gerrit:1283101|VE: Avoid counting all refs when listIndex is undefined (T425433)]], [[gerrit:rCICF12830376bd04|search: fix alt. completion indices to test keyword tokenizer (T420427)]], [[gerrit:1283041|search: enable Latin-to-Devanagari transliteration second-chance (T425018)]] (duration: 08m 58s)
Comment Actions
The patch got deployed but the example in the parent task does not yield the expected results.
tendua seems to be translated to तेंदअ which is close to तेंदुआ and thus probably relies on fuzzy matches moving leopard very far down in the ranking (not visible in the top 10)
Comment Actions
I forgot about the original query in the ticket! It didn't make it into the top ~1400 most common query words so it isn't hard coded.
The final characters अ and आ differ by vowel length (a vs aa). People seem not to transliterate long vowels at the ends of words. Latin n is also hard to transliterate correctly, and the transliteration got that wrong, too.
However, my hope is that when people see the transliteration in action they will figure out how to get better results from it, as tenduaa gets the correct result in the third spot. Doubling the final vowel when needed is definitely the thing most likely to improve transliteration results (if the partial word didn't already match something useful).
Looking at the suggestions for tenduaa, you can see why this is so hard to begin with! There's तेन्दुआ (the leopard) and तेंदुआ (the town). Google translate transliterates both as tendua. Oh, and I see that both are "leopard" as well, with one being a redirect to the other—so we actually have two chances to match with different spellings for the "n" if we get the double aa right!