Language Infrastructure Framework

A Reference Model for What It Takes for a Language to Live End-to-End Online
‍
The Internet Isn’t a List of Features. It’s an Experience
‍
Domain names, websites, and apps all matter but users don’t experience the internet in pieces. They experience it as one continuous journey: expressing ideas, making decisions, navigating systems, and managing their lives online.

True digital inclusion means language is present at every step. When a language is readable but not writable, or available for browsing but missing during payments or safety warnings, users are forced to abandon their language midstream.
‍
The Language Infrastructure Framework maps where language must exist end-to-end so everyone can participate online in their own language.

‍The Layers of the Language Infrastructure Framework

Our Core Benchmark: End-to-End Presence

‍A language is only truly present online if someone can complete everyday digital tasks entirely in that language without being forced into an unwanted fallback.

To achieve this, a language must progress through three interconnected phases:

Foundations: Systems can reliably process, render, and recognize the language.
Interaction: Tools, interfaces, and inputs work seamlessly in-language.
‍Ecosystem: Information, services, algorithms, and safety protections exist for everyone.
‍

A Diagnostic Tool, Not a Scorecard
‍
By examining the user journey layer by layer, we can pinpoint exactly where a language drops out and where exclusion builds up. The goal isn't to rank or compare languages, it's to establish a shared vocabulary so our global community can target structural barriers and build lasting solutions.
‍

Technical Foundations

Layer 1: Encoding and Script Legibility

Before anything else can work, a language must be legible to machines. Its script needs to be encodable, renderable, and stable across the environments people actually use. When this layer fails, users encounter broken characters, missing glyphs, or substituted symbols. At that point, the Internet cannot reliably see the language. This layer is often treated as solved, but in practice, it depends on consistent implementation across platforms, updates, and device classes.

In scope of this layer:
‍

Unicode encoding and normalization
Font availability and fallback behavior
Rendering across major operating systems
Rendering across modern web browsers
Display consistency across desktop and mobile devices

An analysis at this level would typically involve reviewing standards bodies, OS and browser implementations, font libraries, and real-world rendering behavior across common devices.

If this layer breaks, every subsequent layer inherits the failure.
‍

Layer 2: Identity and Addressability

Once a language can be rendered, it must be usable for identity. People need to be able to name themselves and be reached in their own language through domain names, email addresses, URLs, usernames, and form fields.

This is where Universal Acceptance sits. If systems reject a language at this layer, users are forced to transliterate, simplify, or abandon their language at the moment they establish an online presence. Inclusion that fails at identity is fragile by definition.

In scope of this layer:
‍

Internationalized domain names (IDNs)
Email address internationalization (EAI)
URLs and link handling
Usernames and account identifiers
Form validation and input acceptance
‍

Layer 3: Input and Expression

An analysis at this level would typically involve registries, registrars, email providers, DNS operators, platform login systems, and common form-validation behavior across major services.
‍

Even when a language can be displayed and accepted by systems, users still need to be able to express themselves. This is where many languages quietly fall out of the Internet experience.

Input is not just about keyboards. It is about whether people can reliably type, compose, and correct text using the tools available to them across devices and everyday contexts. A language that can be read but not written is functionally muted. It exists online, but only as something spoken to users, not by them.

In scope of this layer:
‍

An analysis at this level would typically involve operating system input support, mobile platforms, IME availability, and observation of how users actually create text in real-world settings.

This is often the first point where users adapt themselves to the Internet rather than the Internet adapting to them.

Physical keyboard layouts
Virtual and mobile keyboards
Input Method Editors
Predictive text and basic input aids
Voice input where relevant
AI-assisted text composition and correction tools

User Interaction with Systems

Layer 4: Core Tools and Applications

For a language to be usable, it must function in the basic tools people rely on to exist online. This includes not just consuming content, but communicating, authenticating, and understanding system responses.

The question at this layer is not whether the experience is polished, but whether it is intelligible enough for users to act with confidence. Partial support here where some elements appear in-language, and others revert to a default, creates constant mental translation and quiet exclusion.

In scope of this layer:

Web browsers including menus, dialogs, and settings
Email clients including composition, reading, and system messages
Messaging applications
Authentication and account management flows
System notifications and alerts

An analysis at this level would typically involve mainstream browsers, email clients, messaging platforms, and account systems, with attention to default language behavior and mixed-language flows.

This layer defines whether a language is merely present or actually usable.
‍

Layer 5: Navigation and Interfaces

As users move through the Internet, they rely on interface elements to orient themselves and make decisions. Menus, settings, consent dialogs, error messages, and help text shape whether users understand what the system is asking of them and feel confident acting on it.

When interface language is inconsistent or missing, users guess. They click through without understanding. Over time, this erodes trust and agency. Inclusion at this layer is less about linguistic polish and more about reducing friction and fear.

In scope of this layer:

Menus and navigation elements
Settings and preferences
Consent dialogs and notices
Error messages and system feedback
Help and support text

An analysis at this level would typically involve reviewing user interfaces across major platforms and services, especially points where users must make decisions or respond to system prompts.

Participation in the Broader Internet

Layer 6: Content Availability

Technical support without content leads to an empty experience. For a language to feel present, there must be something to engage with, such as information, education, culture, and community.

This does not require exhaustive coverage, but it does require suﬃcient breadth and relevance for users to feel that their language belongs online. People need to be able to publish information in their own language.

In scope of this layer:

News and informational content
Educational and reference material
Cultural and community resources
Public-interest information
Locally relevant digital content

An analysis at this level would typically involve surveying major content platforms, content management systems (CMSs), public information sources, and community-driven spaces to understand what is actually available in-language.
‍

Layer 7: Algorithmic Mediation and AI Systems

Even when content exists and is technically accessible, users increasingly encounter it through algorithmic systems. Search engines, recommendation systems, automated translation tools, generative AI assistants, and large language models now mediate how language is interpreted, summarized, surfaced, and understood.

This layer determines not only whether content can be found, but whether it is accurately represented, culturally understood, and usable. In many contexts, AI systems now replace direct navigation entirely, synthesizing answers rather than pointing to sources. Those AI systems are first trained on data that was found on the Internet.

When languages are underrepresented in training data, poorly modeled, or treated as low-confidence by AI systems, the result is subtle but powerful exclusion. Content may exist, but it is misinterpreted, deprioritized, summarized incorrectly, or ignored entirely. Over time, this makes a language feel digitally marginal, even if technical infrastructure is in place.

In scope of this layer:

Search indexing and ranking
Query understanding and intent matching
Recommendation systems
Machine translation quality
Speech recognition and synthesis systems
Large language model training data representation
Generative AI summaries and assistant responses
Automated content moderation models

An analysis at this level would typically involve:

Testing in-language queries in search and AI assistants
Evaluating translation and summarization fidelity
Reviewing dataset representation in training corpora
Assessing whether culturally specific knowledge is preserved or distorted
Examining how AI confidence scoring aﬀects ranking and surfacing

Failures at this layer signal that a language may be technically supported, but algorithmically diminished.
‍

Layer 8: Transactions and Participation

Participation goes beyond reading and browsing. People need to be able to act by submitting forms, completing transactions, accessing services, and participating in civic and commercial life in their own language.

When languages disappear at moments of consequence, users are included only at the surface level. They are present, but not empowered.

In scope of this layer:

Forms and submissions
Payments and commerce
Government and civic services
Account changes and confirmations
Meaningful online participation

An analysis at this level would typically involve service platforms, e-commerce systems, and public-sector interfaces where language directly aﬀects the ability to act.
‍

Layer 9: Trust, Safety, and Accountability

Finally, language must exist where risk, responsibility, and harm are involved. Warnings, security notices, moderation messages, and appeal processes determine whether users can protect themselves and understand their rights.

Absence at this layer does not just inconvenience users. It increases vulnerability.

In scope of this layer:

Security and fraud warnings
Scam and abuse notices
Moderation and enforcement messages
Appeals and dispute resolution
Terms and policies that materially aﬀect users

An analysis at this level would typically involve trust and safety systems, including automated and AI-assisted moderation workflows, and user protection mechanisms across platforms and services.
‍