Summary

  • Source-language choice is an allocation of agenda power: it determines who can draft directly, which concepts appear natural and who must react through translation.
  • Translation quality should be measured against the affected language population, document lifecycle and decision window, not only the number of versions posted.
  • Institutions should separate authoritative text, working text and access translations; publish terminology decisions, version relationships, unresolved ambiguity and the legal effect of discrepancies.
  • Multilingual governance begins with co-design, multilingual concept collection and review before a first draft, followed by simultaneous release, equal comment routes and evidence that non-English contributions changed the text.

The first sentence is already a decision

By the time a policy is sent for translation, its problem, actors, verbs and available remedies have been chosen. The source text may distinguish allocation from assignment, member from entity, advice from consent, or shall from should. Translators can render those distinctions carefully. They cannot recover concepts that the drafting group never considered.

This is the power exercised before translation begins. People fluent in the source language can propose text, negotiate ambiguity and use familiar institutional terms in real time. Others receive a completed frame and are asked whether the translation is accurate. Their role is correction rather than authorship.

English is the working language across major Internet governance venues. That common medium reduces coordination cost among entities from many language communities. It also makes English-language drafting practices appear neutral. They are not. Every language carries legal traditions, technical conventions and assumptions about agency.

The remedy is not to pretend every document can be drafted independently in every language. It is to recognise source-language choice as governance and distribute conceptual input before wording hardens.

Count affected language communities, not posted files

An institution that publishes six translations may report broad multilingual access. The stronger denominator is the population expected to understand, implement or contest the decision and the languages in which those people can participate effectively.

For a global policy, report the language distribution of members, operators, meeting entities, commenters and affected users where data can be collected lawfully. For a regional registry, compare service-region languages, member preferences, policy-list participation and survey responses. No list of six or eight languages can represent Asia Pacific or the world completely.

Document counts also hide importance. Translating a homepage and annual summary does not equal translating the proposal, rationale, impact assessment and amendment instructions that determine rights. Coverage should be weighted by function and decision stage.

The denominator for timeliness is the full comment window. A translation posted twenty days into a thirty-day consultation gives its readers one-third of the practical opportunity available to source-language readers. “Available before close” is not equality.

Translation access and drafting power are different

Translation access asks whether a person can understand a text in another language. Drafting power asks whether that person’s concepts and preferred distinctions can shape the source from which all versions flow. Good translation can succeed at the first while leaving the second untouched.

Suppose an English draft describes address resources as assets, while another language’s operator community commonly understands them as conditional administrative rights. A faithful translation of “asset” may intensify the original assumption. The translator should flag the conceptual conflict, but the policy group must decide it.

Multilingual review therefore needs two questions. Does the target version accurately express the source? Does the source itself express the relevant concept across the communities it will govern? The second cannot be assigned solely to language staff.

Institutions should create a route for target-language reviewers to reopen source wording when it does not travel. This is not translation failure. It is evidence that the original formulation was too narrow or ambiguous.

ICANN acknowledges English as the working language

ICANN’s draft Language Services Policy and Procedures stated that its working language is English while describing translation and interpretation support in the six United Nations languages where appropriate and subject to budget constraints. It covered translation, simultaneous interpretation, teleconference interpretation, transcription and real-time transcription.

The policy recognised multilingualism as essential to a global multistakeholder organisation. That recognition is important. The listed services address document and meeting access at several stages.

The constitutional question remains: what becomes English before it becomes multilingual? If staff papers, community drafts and board materials originate in English, entities using other languages enter at different points. A translated final report cannot equalise earlier negotiation.

ICANN can make the boundary visible by identifying source language, drafting group languages, translation status, authoritative version, discrepancy process and whether non-English submissions changed the source. Multilingual output should be accompanied by multilingual influence evidence.

The IGF interprets main sessions but works mostly in English

The IGF’s official conference FAQ says the working language is mostly English and interpretation is provided in all six UN languages for main sessions. This gives many entities real-time access to prominent discussions. It does not cover every workshop, corridor conversation, preparatory call or drafting exchange.

Main-session interpretation can produce a misleading headline if the programme contains many smaller rooms where substantive networks form. Report interpreted session hours divided by all official session hours, room capacity covered, remote availability and recordings.

The IGF also welcomes volunteer translation of content from English into UN languages and notes machine translation limitations. Volunteer energy expands reach, but dependence on volunteers can create uneven timing and quality. Institutional responsibility should match the importance of the document.

The crucial direction is visible: translation proceeds from English source. A stronger process would collect thematic input in multiple languages before themes, messages and summary categories are fixed.

APNIC's regional experience shows demand is uneven

APNIC has long confronted a service region with extraordinary linguistic diversity. A 1999 independent survey reported strong interest in translating policies and procedures, with differing views about whether APNIC, economies or national registries should be responsible and whether versions should be simultaneous.

APNIC later translated policy proposals into eight regional languages around APNIC 40 and reported translated outcomes as a regular feature. This addresses a consequential layer: proposals that can change number-resource rules.

Survey evidence also shows why a single aggregate can mislead. APNIC’s 2014 survey found that most respondents did not report English as a barrier, while a substantial share of those who did were from developing economies. A minority overall can still contain communities for whom the barrier is concentrated and consequential.

Language investment should therefore use disaggregated need, policy importance and risk, not majority comfort alone. The people most able to answer an English survey are not a neutral denominator for deciding whether English excludes participation.

The IETF makes the common-language trade-off explicit

The IETF states that English is its official language. Its diversity page also notes survey evidence that many entities who use English as a second language find written English easier than spoken English, and it provides closed captioning for official meeting sessions.

Captioning can reduce auditory and second-language barriers. Written archives allow slower interpretation and external translation. These measures improve access within an English-governed process.

Standards drafting depends on exact, shared technical language. Multiple independently authoritative versions could create interoperability risk. That operational need does not remove the governance effects of English fluency: who authors drafts, speaks quickly, chairs discussion and detects nuance.

The IETF can measure those effects through survey, authorship, chair and speaking data while preserving English as the authoritative standard language. Multilingual explanatory materials, terminology and pre-meeting support can widen entry without creating competing protocol requirements.

Concept selection happens before word selection

Translation debates often focus on equivalent words. The earlier decision is which concept deserves a word. A draft may frame a dispute as efficiency, scarcity, abuse, property, security or participation. Each frame directs evidence and remedies differently.

Multilingual concept collection should occur before drafting. Ask affected communities, in their chosen languages, how they describe the problem, which distinctions matter, what examples reveal failure and which terms carry legal or cultural consequences. Language professionals and subject experts can synthesise the results.

The first outline should record competing frames rather than silently choose one. Drafters can then explain why a concept was included or rejected. This makes later translation more accurate because translators know the intended construct.

Without this step, translation becomes a one-way delivery service. The source language supplies the theory; target languages supply labels.

Definitions are sites of institutional power

A definitions section can appear technical while deciding membership, eligibility, rights and enforcement. Terms such as community, consensus, operator, user, public interest and reasonable carry contested meanings even within English.

Every defined term should have a concept note: intended meaning, excluded meanings, examples, legal or technical references and known translation risks. Multilingual reviewers should test whether the boundary survives in their contexts.

If one target language requires two terms where English uses one, that may reveal an ambiguity worth repairing in the source. If several languages collapse a distinction, examples or separate labels may be necessary. The source should not automatically prevail merely because it came first.

Terminology decisions need version history. A changed term can alter past interpretation and implementation. Publish why it changed and which documents are affected.

Grammar can move responsibility

Passive constructions hide actors: “resources were revoked,” “consensus was found,” “concerns were addressed.” Some languages require different choices about subject or agency. A translator may have to identify who acted when the source avoided doing so.

Policy drafting should prefer explicit actors, duties and conditions. State which body allocates, reviews, decides, appeals and enforces. This improves every language and makes authority auditable.

Modal verbs are equally consequential. Shall, must, may, should and will do not map perfectly across legal traditions. A terminology note should specify obligation, permission, recommendation or prediction.

Ambiguity should not be outsourced to translators. Where the governing body intentionally preserves discretion, it should say who holds it and how it is reviewed.

Metaphors import a political model

Internet governance relies on metaphors: ecosystem, community, bottom-up, stewardship, marketplace and multistakeholder table. They make complex institutions legible while hiding boundaries. A “community” sounds voluntary and equal even when voting and contracts are concentrated.

Metaphors may not travel. A literal translation can sound unnatural, while a culturally familiar alternative may carry different authority. Translators need permission to explain rather than reproduce imagery mechanically.

Important metaphors should be paired with operational definitions. Who can join the community? Who sets the table? What power does a steward hold? Which decisions move upward from whom?

If the institution cannot answer those questions, multilingual review has exposed a governance weakness, not a stylistic problem.

Proper nouns and acronyms can disguise institutional assumptions

Technical communities use dense acronyms: RIR, ASN, DNS, PDP, MAG and many more. Keeping official forms can support cross-language recognition. Surrounding explanation must still be accessible.

Glossaries should preserve canonical acronyms while defining role, power and relationship in each language. Transliteration alone is not explanation. Local experts should review whether examples fit actual practice.

Organisation names require care. An official name should not be casually translated into a name the institution does not use. Geographic descriptors and ordinary explanatory text, however, should be localised rather than left as English fragments.

Terminology consistency supports search and participation. It should not force every language into English word order or erase locally meaningful distinctions.

Interpretation changes the timing of participation

Simultaneous interpretation reduces delay but still introduces cognitive and procedural costs. The interpreter must process specialised terms, the listener receives a slight lag, and fast exchanges may move before a response is ready. Jokes, acronyms and overlapping speech make the burden worse.

Chairs should slow pace, avoid interruption, provide materials early and pause after complex questions. Speakers should identify acronyms and use explicit sentences. Remote interfaces must let entities choose language without losing the speaking queue.

Measure floor-language speaking time, interpreted interventions, response rates and interruptions. Do not evaluate interpreter performance from entity silence alone; meeting design may be the limiting factor.

Interpretation coverage should include the sessions where decisions are shaped, not only ceremonial plenaries. If budget requires prioritisation, publish the criteria and unmet demand.

The first draft receives the longest review

Source-language readers often see early drafts, mailing-list discussion and amendments. Target-language readers may receive only a consultation version. Even simultaneous formal publication cannot compensate for months of source-language negotiation.

Map document history from issue identification through outline, drafts, committee review, consultation, decision and implementation. At each stage, record languages available and routes to contribute. Compute exposure time by language.

Where full translation of every draft is impractical, provide multilingual issue papers, change summaries and structured input before the first source draft. Translate consequential amendments and extend deadlines when target versions change materially.

The goal is not identical paperwork. It is equivalent opportunity to shape the decision before costs of change become high.

Translation lag is an authority gap

A policy may technically remain open for comment while staff and regular entities begin converging around the source. Late translations enter a discussion whose social decision has advanced. Readers can submit, but their chance of changing text is smaller.

Report median and maximum lag for each document class. The denominator is hours or days from source release to consultation close. Set service targets based on risk and importance.

If a translation is delayed, pause or extend the common deadline rather than grant only the affected language a separate window after others have moved on. All entities should see cross-language comments and responses before closure.

Urgency exceptions should be narrow, documented and reviewed. Repeated emergency English-only publication indicates planning failure.

Machine translation changes cost, not accountability

Machine translation can make large volumes accessible quickly. It is useful for discovery, preliminary understanding and low-risk material. It can also produce fluent errors that readers cannot easily detect, especially in technical or legal text.

Every machine-generated version should be labelled with status and date. High-impact documents require human subject review. Readers need a route to report errors and see corrections.

Do not use machine output to claim equal language status when only the source is authoritative and professionally maintained. Coverage metrics should distinguish raw machine access, reviewed translation and authoritative text.

Automation can free human capacity for conceptual review if savings are reinvested. It should not turn target-language communities into unpaid quality assurance.

Volunteer translation is valuable but structurally fragile

Volunteer communities bring expertise, trust and languages institutions may not otherwise support. They can adapt terms to local practice and distribute knowledge. Their work should be credited, supported and reviewed.

Critical obligations cannot depend entirely on spare time. Volunteer capacity varies, deadlines slip and quality burdens fall unevenly. The institution remains responsible for accuracy and decision timing.

Offer terminology tools, subject briefings, paid review for high-risk texts and clear licensing. Do not require public attribution where safety is a concern. Preserve independence so volunteers can challenge the source.

Measure unpaid hours where entities consent. Language inclusion should not be achieved by externalising institutional cost to multilingual members.

Who decides which version is authoritative

Every multilingual policy needs a status rule. One version may be legally authoritative, several may have equal authority, or translations may be informational. Ambiguity becomes dangerous when versions diverge during a dispute.

Publish the rule prominently on each document, not in a remote policy page. Explain how discrepancies are reported, who interprets them, whether enforcement pauses and how corrections affect earlier actions.

If English controls, the institution should still treat a translation error as its own failure. A person who reasonably relied on an official version should have a remedy, especially where rights or deadlines are affected.

Equal authority requires a conflict rule. Courts and multilingual legislatures have methods for reconciling versions; Internet institutions should not improvise only after a dispute.

Comments must cross languages, not remain in lanes

Accepting comments in multiple languages is limited public evidence if decision-makers read only an English summary. Translation must operate in both directions. Source-language entities need access to non-English contributions, and the contributor needs to verify the translated meaning.

Publish original and translated comment together with consent and privacy controls. Allow correction of the translation. Staff summaries should identify which language raised an issue and how it affected analysis without treating language as a proxy for region or view.

Response tables should address substantive points equally. Count comments, unique contributors, issues accepted, issues rejected with reasons and text changes by submission language. Small numbers require caution.

The strongest evidence of multilingual participation is not submission volume. It is traceable change and reasoned response.

Consensus can be language-sensitive

Rough consensus and chair judgment depend on hearing intensity, objections and alternatives. Entities operating in a second language may speak later, more briefly or less often. Silence can be mistaken for assent.

Chairs should invite written follow-up, restate decisions plainly, confirm objections across language channels and keep the record open long enough for interpreted review. Consensus calls should identify unresolved issues, not only room mood.

APNIC’s policy process considers mailing-list, meeting and remote participation. Multilingual proposal and outcome translation can widen understanding, but consensus assessment should also examine whether comments in regional languages entered the chair’s reasoning.

No language group should be counted as supportive merely because few objections were expressed in the working language.

Elections and membership notices carry special risk

Nomination deadlines, candidate rules, voting instructions, fees and bylaw changes directly affect member rights. These documents should receive priority translation and simultaneous release where the institution serves multilingual members.

The audit should compare notice time, open and click rates only as rough access indicators, questions received, rejected ballots and support requests by language. A translated reminder cannot cure an English-only nomination rule that candidates encountered too late.

Candidate statements may be translated under equal limits and review. Live forums need interpretation or equivalent recorded answers. Discrepancies should not advantage one campaign.

If a language error affects eligibility or voting, the remedy may require deadline extension, ballot correction or rerun. Apology alone is not accountability.

Implementation can diverge after accurate translation

A policy can be translated correctly yet implemented through English-only forms, support channels and training. Operators then face a different rule in practice. Language governance must follow the full service path.

Test application instructions, interfaces, error messages, contracts, help desks, appeal notices and implementation guidance. Record which versions are maintained when policy changes. A stale translated form can be more harmful than no translation because it appears official.

Support staff need terminology and escalation routes. Questions that expose ambiguity should feed back to policy owners. Repeated issues by language may reveal a source-text defect.

Implementation audits should include users who do not attend meetings. Conference participation is a poor denominator for operational access.

A multilingual drafting protocol

Before drafting, publish the issue in plain language across selected languages and collect concepts, cases and terms. Name the language-selection method and gaps. Form a drafting group with multilingual subject expertise, not language staff alone.

Create a concept map and terminology register. Mark contested terms. Produce a source draft with explicit actors and duties. Translate early enough for target-language review to change the source. Release consultation versions simultaneously and translate comments both directions.

Record version relationships, authoritative status and unresolved discrepancies. After decision, translate implementation materials and monitor questions. Review whether non-source-language input changed definitions, scope, evidence or remedy.

The protocol turns translation from a final production step into a recurring governance function.

Denominators for a language-power audit

Report preferred-language members divided by all members with known preference; contributors by language divided by eligible contributors; translated document days divided by source document days before decision; interpreted official session hours divided by all official session hours; and reviewed high-impact translations divided by all high-impact translations.

Also report non-source-language issues receiving responses, source text changes traceable to those issues, translation corrections, implementation support cases and appeals affected by language. Counts and methods belong beside rates.

Unknown language preference must remain visible. Do not infer language from country or name. Self-identification and data minimisation are essential.

The scorecard should separate access, quality, timeliness and influence. A high document-coverage rate can coexist with no drafting power.

Independent review needs bilingual evidence

An English-only audit of multilingualism repeats the problem. Review teams should include independent bilingual or multilingual subject experts and sample both source and target communities.

Test terminology, document timing, meeting recordings, comment treatment and implementation. Interview people who stopped participating, not only successful regulars. Publish methods and limitations in relevant languages.

The institution should answer findings across languages and assign owners and dates. Corrections must propagate to every maintained version.

Reviewers should not rank languages by entity worth. Prioritisation can reflect risk, affected population and demand while acknowledging uncovered communities.

What equality can mean when one text must control

Some technical and legal settings require one authoritative version. Equality then cannot mean identical legal status. It can mean equal opportunity to shape the controlling text, timely access to understand it, reliable interpretation, a route to correct errors and a fair remedy for reasonable reliance.

This model is demanding but practical. It preserves interoperability and legal clarity while refusing to treat English fluency as consent. It directs resources toward early concept work and two-way participation rather than counting final PDFs.

Institutions should say openly why one version controls. Hidden hierarchy breeds more distrust than declared hierarchy accompanied by meaningful safeguards.

The test is whether language communities can alter the controlling text before it binds them.

Language budgets reveal the institution's risk model

Budget decisions show which audiences the institution expects to bear misunderstanding. Translation, interpretation and review are often treated as communications costs that can be reduced when finance tightens. For policies affecting rights, resources or elections, they are governance controls.

Publish language spending by function: document translation, interpretation, review, terminology, accessibility, platform and support. Compare it with total policy, meeting and engagement expenditure. A rising number of machine-translated pages should not be reported as efficiency if professional review of consequential text declines.

Prioritisation criteria should include legal effect, operational risk, affected language population, decision window and evidence of demand. Ceremonial visibility should not displace voting instructions or implementation guidance merely because it produces better photographs.

Boards need unmet-demand reporting. Which languages and document classes were requested but not supported? What harms or participation losses followed? Budget constraints are real; hidden rationing is a choice without accountability.

Legal traditions do not align word for word

Terms such as ownership, licence, trust, appeal, due process and public interest carry legal histories that differ across jurisdictions. A linguistically fluent translation can still import the wrong legal relationship if the reviewer lacks subject expertise.

High-impact policies require paired language and legal review. The concept note should state whether a term invokes a particular jurisdiction’s law or functions as an institutional term. Where no equivalent exists, explanation may be safer than a familiar but misleading label.

Contracts and bylaws need a conflict rule and notice of controlling law. Members should know whether a translated version has legal effect and what happens when reliance causes loss. Deadlines should pause where the institution corrects a material error.

Legal review must not turn every target version into a paraphrase of English doctrine. Local experts should be able to identify where the source assumes a legal category that is inappropriate for a global technical institution.

The translation memory can preserve yesterday's mistake

Institutions use terminology stores and translation memories to keep wording consistent. These tools improve speed and coherence. They can also reproduce an outdated or contested term across hundreds of documents after the underlying concept changes.

Every approved term needs owner, definition, context, date, reviewer and status. Deprecated terms should remain searchable with an explanation, not disappear. Material changes need propagation across active documents and interfaces.

Language communities should have a visible correction route. Repeated objections to one term require concept review, not automatic rejection because the memory marks it approved. The institution should publish significant terminology decisions and unresolved variants.

Version control is particularly important where one acronym maps to different official terms across regions. Consistency should help readers trace the institution, not erase regional usage.

Cross-language disputes are valuable test cases

When entities disagree about a translation, the institution should preserve the case as governance evidence. Record the source wording, target wording, claimed ambiguity, affected action, decision-maker, correction and remedy. Anonymise personal details where needed.

Annual analysis can identify recurring problems: modal verbs, role names, deadlines, resource rights or appeal instructions. These patterns should shape drafting guidance and budget. One material dispute may justify more attention than thousands of error-free marketing pages.

Do not treat every complaint as a translator error. The source may be ambiguous, the governing rule may be inconsistent, or different language communities may expose a policy conflict. The resolution body needs authority to send the issue back to policy owners.

Publishing de-identified case summaries builds trust and helps operators understand how discrepancies are handled. It also gives reviewers a denominator beyond raw document counts.

Multilingual minutes determine institutional memory

Live interpretation may allow participation, but the lasting record is often a source-language transcript, minutes or summary. If objections expressed through interpretation are compressed or mistranslated, later decision-makers inherit a different meeting.

Entities should be able to review the record of consequential interventions in the language used and in the institutional working language. Corrections need timestamps and visible history. Audio should be retained under clear rules long enough to resolve disputes.

Minutes should identify disagreement, conditions and unresolved terms rather than convert debate into smooth consensus. Summaries translated after approval should receive the same conceptual scrutiny as proposals because they become evidence for later action.

The denominator for multilingual participation must therefore include preserved interventions, not only interpreted minutes of live speech. Access that vanishes from institutional memory has limited governing effect.

Local-language meetings can feed the global process

National and regional operator groups, Internet governance initiatives and member events often discuss issues in languages absent from global mailing lists. Their conclusions should not enter only as English summaries written by a central intermediary.

Provide a structured route for local groups to submit original-language records, translated key points, evidence and disagreement. Fund translation where the global institution requests input. Let contributors verify the summary before it is cited.

Do not treat one local meeting as representative of a country. Publish attendance, selection, process and limits. The value lies in specific operational evidence and concepts, not geographic certification.

This federated approach can distribute early framing power without requiring every entity to join one English-language venue. It also reveals where a global term has different practical meanings across networks.

Search and discovery can create a second language hierarchy

A translated document may exist and still remain practically invisible. Search engines, site navigation, document titles and internal links often privilege the source version. Users searching local terminology may never find the official text.

Audit indexation, language tags, canonical links, sitemaps, on-site search and cross-language navigation. Each version should identify its status and link to all maintained languages. Search results should not replace a reviewed translation with an automatically generated English excerpt.

Measure discovery from local-language queries to consequential documents, not only page publication. Support logs can reveal when users repeatedly ask for material that technically exists. Fix terminology and navigation rather than blaming search behaviour.

Archive policies need parity too. If an old source remains accessible while its translations disappear during a site migration, historical accountability becomes language selective. Preservation plans should include every official version and correction history.

Data and charts require language governance too

Charts can appear language neutral because numbers remain the same. Titles, units, category definitions, notes, colour meanings and suppressed values determine interpretation. A translated caption cannot repair a dataset whose categories were defined only in the source language.

Publish multilingual data dictionaries for consequential statistics. Keep machine-readable identifiers stable while localising reader labels. Explain rounding, missingness and exclusions in each language. Alternative text and table versions are necessary for accessibility.

When respondents answer in several languages, analysts should test whether categories have equivalent meaning before combining results. A shared numeric scale may conceal different interpretations of “satisfied,” “member” or “operator.”

Dashboards must preserve active language through filters, exports and error states. Falling back to English at the point of detailed analysis gives source-language readers a deeper evidence surface than everyone else.

Emergency notices expose the real language priority

Security incidents, election corrections and service outages create pressure to publish quickly. Institutions often issue a source-language notice first and promise translations later. The delay can affect operational action, rights and safety.

Prepare approved terminology, templates and interpreter rosters before emergencies. Identify languages tied to affected services and publish verified core instructions simultaneously where feasible. A short accurate notice is better than a long source text with uncertain automated copies.

After the event, report release times by language, corrections, reach and consequences. Emergency exceptions should feed capacity planning. If the same languages repeatedly wait, the exception has become policy.

Urgency can justify temporary asymmetry; it cannot erase accountability for who bore the risk. Remedy may include deadline extensions, direct outreach or reconsideration of actions taken under incomplete notice.

Conclusion: multilingualism begins with the blank page

ICANN’s language services, IGF interpretation, APNIC’s regional translations and IETF captioning all reduce real barriers. They make documents and meetings more accessible, and they deserve serious investment. But they operate inside processes where English frequently defines the first frame.

The first drafting language chooses categories, allocates authorship time and determines which entities react to a finished structure. Translation can reproduce that structure brilliantly without sharing the power that made it. A count of versions therefore cannot establish multilingual legitimacy.

Begin earlier. Gather concepts and cases in multiple languages. Track the full document lifecycle. Let target-language review reopen source wording. Publish authority and discrepancy rules. Translate comments both directions. Measure whether non-source-language evidence changes text, decisions and implementation.

The blank page is the point at which language equality is cheapest and most powerful. Once the first draft becomes institutional common sense, every translation starts from behind.

Institutions should treat the resulting friction as useful evidence rather than inefficiency. A concept that cannot be expressed clearly across the communities expected to implement it may be underspecified. A target-language reviewer who reopens the source is improving the rule for everyone. Multilingual governance is not the production of parallel documents after agreement; it is a method for testing whether agreement rests on concepts that entities actually share.

Sources