Wikilambda and the Old Dream of a Perfect Language

The Wikimedia Foundation does not raise venture rounds or chase market share, but it does ship ambitious infrastructure, and its newest bet is one of the strangest in the open-source world. Wikifunctions, launched in 2023, is a collaboratively edited library of computer functions. Sitting underneath it is Abstract Wikipedia, a project running since 2020 that wants to store knowledge in a language-independent form and then generate readable articles in any human language on demand. The plumbing that connects the two is an extension to MediaWiki called Wikilambda.

A new paper by Michael Falk of the WikiHistories project, published in AI & Society, looks hard at that plumbing and reaches an uncomfortable conclusion. Wikilambda, Falk argues, is the latest entry in a very long line of attempts to build a 'perfect language,' and it inherits the same structural reasons those attempts have failed before.

{{IMAGE:2}}

The problem Abstract Wikipedia is trying to solve

Start with the actual problem, because it is a real one. Most of the world's roughly 300 Wikipedia language editions are thin. English has nearly seven million articles; many languages have a few thousand. Writing and maintaining encyclopedic coverage separately in every language is a task no volunteer community can finish. Abstract Wikipedia proposes a different path: store a fact once as structured data drawn from Wikidata, then run functions that render that data into a grammatical sentence in whatever language a reader wants. Write the renderer once, serve every language.

Wikilambda is the system that makes those renderers runnable and editable by anyone. It treats each piece of a computation not as a named keyword but as a database entry with an identifier, a Z-number or Z-key. When you open a function in the interface, those identifiers are displayed in your preferred language. The pitch is that programming itself can be unhooked from English.

Two flavors of an old idea

Falk's analytical move is to borrow Umberto Eco's 1995 book The Search for the Perfect Language, which catalogs historical attempts to design ideal languages. Eco separates two ambitions. A perfect language mirrors the true structure of reality, each word mapping to a real component of the world, each rule mapping to how the world combines those components. A universal language is one everyone can or should speak. Esperanto is the famous universal attempt among human languages; among programming languages, BASIC, Logo, Python, and Scratch carry similar accessibility goals.

Wikilambda, Falk contends, reaches for both at once. The template language for Abstract Wikipedia is meant to be perfect, able to express any fact, and universal, usable by writers everywhere. And the history Eco documents is mostly a history of failure. Esperanto never became the lingua franca its logic promised. As one cited study of medical-informatics standards put it, a proposed standard, no matter how clean and well designed, often struggles to displace an imperfect but functional system that already works. English, with all its inconsistencies, won the role Esperanto was engineered for.

Where the argument bites

The most interesting part of the critique is about contradiction rather than difficulty. Wikilambda's designers justify the Z-number scheme as a way to break what they call the hegemony of English and to avoid reproducing Western, imperialist assumptions baked into language. Fair enough. But Falk points out that this rests on a view of language as culturally loaded and irreducible, which sits awkwardly next to the project's working assumption that language is a simple conduit, a container you pack facts into and a reader unpacks at the other end. You cannot easily hold both beliefs. If meaning is just transport, the cultural argument for abolishing English keywords weakens; if meaning is cultural and metaphor-laden, the dream of clean language-independent facts gets harder.

There is also a practical wrinkle. Because English functions as the de facto shared tongue, the developer community building these multilingual abstractions tends to coordinate in English anyway. The hegemony reappears one layer up.

Falk's close reading of the code finds the same tension in the architecture. He examines the function orchestrator, the component that runs a piece of Wikilambda code and uses an ImplementationSelector to pick among multiple available implementations of the same operation. Ordinary languages give you one way to add two numbers; if there are two, the programmer chooses. Wikilambda lets many implementations of an operation coexist and hands the choice to the orchestrator rather than the author. It is a genuinely novel abstraction, and Falk reads the 'orchestration' metaphor seriously. His finding is that the language keeps reaching for fresh abstractions to escape conventional programming, then falls back on default conventions when the abstraction gets too costly to sustain.

Why this is worth watching

None of this is a takedown. The paper opens by recalling The Signpost's 2023 coverage of an outside evaluation that already flagged the project as at substantial risk of failure, and Falk treats Wikilambda as a serious object deserving serious analysis. The review calls it a good introduction to the whole effort and a convincing map of its likely failure points. The method, Critical Code Studies, reads source code the way a critic reads a text, looking at the metaphors developers choose and what those choices reveal about intent.

The ending lands on an irony worth keeping in mind for anyone tracking how knowledge infrastructure gets built. Wikilambda's team has openly criticized 'one ring to rule them all' designs, the single centralized system that swallows everything, yet the thing they are building looks a lot like one. What sets it apart is a moral commitment that most commercial AI translation lacks: the entire pipeline, the structured data, the functions, the interpreter, is transparent, contestable, and editable by humans. Falk's own summary is the sharpest line in the review. If nothing else, he writes, Wikilambda is a thundering critique of corporate AI hype. For a project that may not reach its stated goal, that is not a small thing to be.

In the same research roundup

The Wikimedia Foundation also published results from the 2025 edition of its annual global survey of Wikipedia readers. Across all 11 surveyed language editions, Google and YouTube remain the most frequently named platforms for finding and learning information, but the survey records a clear jump in ChatGPT use for learning between 2024 and 2025, especially among readers of Arabic, Japanese, Korean, Portuguese, and Russian Wikipedia. ChatGPT also drew some of the highest favorability ratings among non-Wikimedia sources, a data point any builder of knowledge products should sit with.

{{IMAGE:3}}

Reader demographics held steady with prior years. Wikipedia's audience skews young overall, though that varies by edition, with German Wikipedia readers trending notably older. The same release covered a batch of Wikidata research, from extracting place names for cross-linguistic semantics, to a New Zealand project linking a nation's doctoral and master's theses, to a study testing how well large language models extract facts from implicit versus explicit biographical text. Taken together, the issue reads as a snapshot of a community building open knowledge infrastructure while the commercial AI products built partly on its data pull more of its readers each year.