UX Research · PhD Dissertation
Is Wikipedia one community — or 312?
A doctoral dissertation that challenged a two-decade, Anglo-centric assumption in collaboration research — characterizing how the English, French, and Spanish editions of Wikipedia actually collaborate, through three mixed-methods studies.
What this work demonstrates
Context & stakes
The Anglo-centric blind spot
For nearly 20 years, researchers extensively analyzed how people collaborate to build quality articles on Wikipedia — but almost entirely on the English edition. Wikipedia runs in more than 300 language editions, yet nearly every intellectual assumption about how Wikipedians work together rested on an Anglo-centric view.
The bias was baked into the platform itself. Wikipedia was founded in English by two Americans; its technical foundation (MediaWiki), its "five pillars," and even its notion of "reliable sources" (the New York Times, CNN) are English- and US-based. With those foundations shared across every edition, it was easy to assume one universal model of collaboration — and never test whether it actually held elsewhere.
But prior work had already shown the "global consensus hypothesis" to be false: encyclopedic world knowledge is not consistent across languages and cultures. If collaboration models don't generalize either, then researching or designing any language edition from English-based assumptions is, as the dissertation argues, bad design — the kind that can quietly distort how information is produced and shared across the world.
Is Wikipedia a standardized platform with one common model of collaboration — or a set of 312 active language editions, each with its own?
My role
Owned the research end to end
I designed and led the entire dissertation — framing the research questions, selecting the right method for each stage, collecting and analyzing data across three languages, and synthesizing three separate studies into a single comparative model. As a native English speaker who learned French and Spanish, I recruited and trained native-speaker research teams for the French and Spanish data, and we met weekly to reconcile coding and translation so the findings held up in each language. The work was advised by a committee co-chaired by David W. McDonald and Mark Zachry in the department of Human Centered Design & Engineering at the University of Washington.
The program
Three studies, three lenses
Rather than study one edition and generalize, I sequenced three studies so each method interrogated the same question from a different angle: what editors do, how they contend for control, and how they make sense of it themselves. Two of the three deliberately replicated canonical English-Wikipedia models — because testing whether a known model travels is exactly how you find out if it generalizes.
What are talk pages actually used for?
I replicated a well-known behavioral model from Viégas et al. (2007) — an 11-category scheme for the posts editors make on article talk pages — and applied it as a lens across the English, French, and Spanish editions, coding 48 talk pages.
The behaviors were universal; their proportions were not. The same categories appeared in every language, in different quantities. The English edition's supposed norm — the 1:2 article-to-talk-page edit ratio Viégas described — no longer held, and probably never held for French or Spanish. Talk pages had quietly shifted from coordinating edits to debating the subject of the article itself; References to Vandalism, for instance, collapsed from 8.5% to 0.9% as bots took over the work.
Same behaviors, different quantities
Across EN, FR, and ES, editors post the same kinds of things — but how often they do each thing differs by edition. A model built on English behavior cannot be assumed to transfer.
How do editors claim control?
Next I replicated Kriplean et al.'s (2007) model of "power plays" — the moves editors use to claim authority over an article through the discourse of policy (article scope, power of interpretation, prior consensus, legitimacy of source, threat of sanction, and more). Native-speaker teams coded 129 talk pages and roughly 2,050 posts, with a five-layer analysis built to catch anything the original English model might have missed.
Data: dissertation Table 5.5 (percentage of coded talk-page threads).
The model is complete — but the frequencies aren't
No new power plays emerged in any language; the original English model held up. Yet the frequency of each play varied by edition — editors claim legitimate control differently across languages, and the consensus process isn't bound to a single language, size, or maturity level.
One constant did hold: WP:NPOV (neutral point of view) was invoked across all three editions, confirming that Wikipedia's oldest pillar runs deep regardless of language.
How do editors perceive power and consensus?
Finally, I interviewed 34 experienced editors — 13 English, 13 Spanish, 8 French — and ran a thematic analysis, deliberately starting with the smallest (Spanish) edition so its patterns shaped the codes rather than being measured against English.
Editors described strikingly similar conversations — sourcing disputes, translation debates, and conflict they mostly tried to defuse — but diverged sharply on where authority actually lives.
Editors disagree on what drives consensus
In Spanish, administrators (called "librarians") drive consensus — the edition dissolved its Arbitration Committee in 2009, concentrating authority in roughly 68 admins. In English, the most experienced editors guide debates with their expertise and tools like RfC and ArbCom. In French, authority mattered far less than the quality and structure of the argument and its sources.
Translation surfaced as its own force: in French and Spanish, importing content from larger editions spawned entire coordination projects (such as WikiProject:Traduction) — early evidence that language itself generates new social processes, not just new content.
The framework
Adding language to the socio-technical system
Wikipedia is a socio-technical system — an instantiation of both social and technical processes. But my studies kept surfacing a third force those two categories couldn't hold: language. Translation, language structure, clarity, and verbosity were actively shaping how editors debated and reached consensus. So I introduced language as a first-class analytical concept alongside the social and the technical — a lens for reading collaboration in any federated, multilingual platform, not just Wikipedia.
Contributions
What the work established
The dissertation makes three contributions, and argues from evidence that Wikipedia must be understood as a sum of distinct parts rather than an English default.
- Empirical — an evidence-based account of collaboration models in three language editions, showing that findings from English Wikipedia do not automatically generalize to others.
- Methodological — a worked example of multi-site, cross-language research, plus a socio-technical framework extended to treat language as a first-class process.
- Validation — a systematic replication of two canonical English-Wikipedia models across three languages, a direct response to the replication gap in HCI/CSCW.
- Reframed the field's default — made the case that designing or studying any edition from Anglo-centric assumptions is bad design, opening a bridge between sociolinguistics and CSCW.
- Extended beyond Western Romance — a follow-on replication in the Farsi and Chinese editions (non-Latin scripts, different language families) was accepted to CSCW 2021.
Reflection
What I'd note — and do differently
- Scale the language expertise. Most analysis ran through one researcher; native French and Spanish speakers gathered and reconciled data weekly, but deeper native fluency at every step would sharpen the reading of nuance.
- Acknowledge the sample. The models are drawn from a predominantly male editor population — consistent with Wikipedia's demographics, but an unbalanced base to generalize from.
- Go beyond Western Romance. English, French, and Spanish share Latin scripts and a language family; testing linguistic-relativity claims needs editions further from those roots — which is exactly why I extended the work to Farsi and Chinese.
- Map the policy regimes. Each edition categorizes policy and guidelines differently; a dedicated study of how editors value and use policy across languages is the natural next step.
About this dissertation. Comparing Language Communities: Characterizing Collaboration in the English, French and Spanish Language Editions of Wikipedia. PhD dissertation, University of Washington, 2021. Committee: David W. McDonald & Mark Zachry (co-chairs), Jennifer Turns, and Andrés Monroy-Hernández. Figures on this page visualize data published in the dissertation.