Wiki Conference India 2026 · Kochi · 4 September

State of Indic
Language Wikis

Santhosh Thottingal
Principal Software Engineer, Wikimedia Foundation
Opening
Why this talk
“
How healthy is my wiki?
  1. Against our own history — where was each wiki at its best, and where is it now?
  2. Against our peers — wikis with similar demography, not against English.
  3. Both lists — the numbers that celebrate us and the numbers that worry us, side by side.
All data is public and reproducible. Every chart in this deck comes from open APIs and open scripts.
Opening
The family

Wiki does not mean only Wikipedia.
89 Indic wikis, seven projects.

30
Wikipedias
from Odia (2002) to Magahi (2026)
20
Wiktionaries
five of them are larger than their Wikipedias
14
Wikisources
the digital libraries of our languages
+ Wikiquote
+ Wikibooks
+ Wikivoyage
+ Wikiversity
15+ writing systems
Scope: languages of Indic typology, not geography — Sinhala and Nepali are in. Counts exclude closed wikis.
Past
Founding waves

Two founding waves

Odia was first (1 June 2002); Assamese, Nepali, and Punjabi followed within the week.
Past
The newest of us

A 3 months old baby wiki

18 June 2026
Magahi Wikipedia (mag) launched
The 30th Indic Wikipedia.
Past
Content growth

Twenty-four years of building

Cumulative content pages created per wiki (Wikistats).
Past
Size is not community

The largest wikis were not written the same way — bot share of edits

Article count and community size are different facts.
Present · Editors
The headline

Active editors grew for sixteen years.
Then we are −27% since 2020.

Registered editors with 5+ edits per month, summed over all Indic Wikipedias (A person active on two wikis counts twice).
Present · Editors
What the edit counts say

Fewer hands, more work

165
edits per active editor per month (2025)
steady at 100-120 for a decade, 148 in 2024 — fewer people carry more of the wiki, likely with better tooling
−11%
human edits since the 2020-21 peak
editors fell 27%; the work fell far less
Edit records set in 2025
Bengali 630k — its best year ever · Assamese 93k, also a record · Tamil at 90% of its best year
Edits at a fraction of the best year
Sanskrit 8% · Malayalam 27% · Hindi 33% · Kannada 43%
Human edits: registered plus IP editors, per year (Wikistats). Edits per editor: registered edits over active editors (5+ edits per month).
Present · Editors
When each wiki peaked

Peaks cluster in two eras. What were the programs behind them?

Each dot is one Wikipedia, placed at the year its active-editor count peaked.
Present · Editors
Celebrations

Bengali held the line. Assamese is at its all-time high.

Bengali: 2025 activity at 92% of its peak — now the largest active Indic community. Assamese: still setting records in 2024-25.
Present · Editors
The other list

2025 activity as a share of each wiki's best year — what will reverse this?

Not a list of failures — a list of communities whose best years prove what is possible.
Present · Editors
Admin capacity

Administrative capacity is thin at the edges — ten Wikipedias have one or two administrators

10 of 30
Wikipedias have 1-2 admins
the current administrator roster has only one or two people
0
admins on Tulu Wiktionary
28 August 2026 snapshot; Tamil Wikipedia has 31, Bengali 16
A 2025 Wikimedia Foundation study found the same pattern movement-wide: admin cohorts are aging and recruitment is not keeping up — Research: Wikipedia Administrator Recruitment, Retention, and Attrition (2025).
Present · Readers
How we are read

Our readers moved to phones. Our editing is at desktops.

Share of 2025 pageviews from mobile. Urdu's low share comes with a suspicious bot activity
Present · Readers
A break in the series

Recorded human pageviews fell after 2023

Bot detection changed in 2025, and March-August data was revised.
AI answers, search changes, and detection changes may all contribute.
Human pageviews per year, all Indic Wikipedias, millions (Wikimedia Pageviews API). The 2025 bot-detection update makes earlier and later data imperfectly comparable; the chart does not estimate AI's effect.
Present · Readers
Country traffic

Indic wiki traffic crosses borders

34%
of Nepali Wikipedia July traffic is from Nepal
23% is from the United States
44%
of Sinhala July traffic is from Sri Lanka
19% is from the United States
US
traffic is prominent
but rankings can include automated traffic
July 2026, Pageviews API by country. This endpoint does not separate human and automated traffic. Bangladesh and Pakistan are excluded by privacy policy; home-country shares are more useful than small-country rankings.
Present · Readers
What readers want

We are the identity encyclopedias — July's most-read article, per wiki

"India" is the #1 reading category on almost every wiki. Readers come to us for who we are — and to English for everything else.
Translation
The counter-trend · articles translated

Fewer editors, better tools, more articles — translations per year, +184% since 2020

Articles published with Content Translation per year, all Indic Wikipedias (2026 is pacing lower, ~40k annualized).
Translation
300,000 articles

For some wikis, translation is how the wiki grows — published translations relative to current article count

2,612
people have translated into Bengali
the widest participation funnel of any editing workflow
299k
translations published into Indic wikis
12% of the 2.5M global Content Translation total; about 7% were later deleted
Translation
The missed opportunity

93% of it flows from English.
Sibling languages barely translate each other: 2.1%

اردو → سنڌي
Urdu → Sindhi · 1,147 articles
हिन्दी → ਪੰਜਾਬੀ
Hindi → Punjabi · 568 articles
ಕನ್ನಡ → ತುಳು
Kannada → Tulu · 334 articles
It works, but it is just rare. Arabic alone sends more into Urdu (12k) than all Indic languages send each other.
Knowledge gaps
Choose your reference

The "gap to English" is mostly a gap of American actors

Top missing categories against English, weighted by English readership — identical across every wiki in this comparison:
Measuring against English measures your distance from American pop culture. That is not the gap our communities need to close. Measure against peers instead.
Knowledge gaps
Peer gaps · interactive

Gap postcards between neighbours

is missing, compared with
Category names appear in the reference wiki's language — the neighbour describing what it can offer. Data: topictrends gap discovery, coverage snapshot 2026-06.
Sister projects
The quiet counter-trend

While Wikipedias lose editors, the sister projects hold or grow

~40
active editors on Tamil Wikisource
a larger community than most Indic Wikipedias
7×
Kannada Wiktionary vs Kannada Wikipedia
266k entries against 36k articles
Active editors per year, summed per project family. Wiktionary set an all-time record in 2025; Wikisource has grown every year since 2023.
Sister projects · Wikidata
Lexicographical data

Wikidata lexemes across Indic languages — counts are not coverage

The big number is hollow: 83% of the Malayalam lexemes carry no senses, and 80% are nouns — a bulk word-list import, not lexicography. And Malayalam is morphologically productive: a materialized word list is a debated model for it in the first place. Hindi and Urdu lexemes live under "Hindustani".
Sister projects · Wikidata
The closing challenge

Items tagged with country: India are usually labelled only in English — 912,073 Wikidata items; who can name them?

No individual Indian language labels more than 1 in 12 of these items. This measures direct country (P17) = India statements, not all India-related items. Every infobox, search result, and assistant that reuses Wikidata inherits this gap.
Close
Interactive

Find your wiki

01 / 25