Translate | ISO 15704:2019 Enterprise Architecture Models, Business Roles, Process Boundaries and Change Plans

How do you translate an enterprise architecture document without changing who is responsible, what a process does, or whether a planned change has actually happened? ISO 15704:2019 offers a reference base for enterprise modelling and architecture, but multilingual teams still need to explain business processes, stakeholder concerns, organisational roles, system interfaces and transition plans faithfully. This guide is for people translating enterprise architecture diagrams, operating models, ERP implementation records, programme requirements and business-model documentation across languages. It focuses on keeping the meaning of a model intact when the words are changed.

Imagine a diagram labelled ‘target operating model’. A reader in another language takes that to mean the organisation already works this way, so a contractor sends work to a nonexistent department. Elsewhere, ‘process owner’ is translated as the person who performs every step, even though the source assigns governance and accountability rather than execution. These are not small grammar mistakes. They alter the interpretation of an enterprise. The solution is to preserve the model’s status, scope, participants, relationships, lifecycle phase and supporting evidence in the translated material.

At a glance: distinguish a reference architecture from a prescribed organisational structure; keep the organisation’s mission separate from its technical systems; label current and proposed states; preserve actor, owner and approver distinctions; trace process triggers and outputs; make external interfaces visible; keep assumptions and version history; and obtain approval before a translated model is used as an operational instruction. The cases below give concrete repairs and a bilingual review workflow without turning this specialist guide into a competing translation master.

Quick answer: a translation that retains the source decision

1. Find the document owner, edition, intended audience and exact purpose of the translation.
2. Separate the model or project description from actual operational authorisation.
3. Map each source term to its concept, evidence and applicable context before choosing target-language wording.
4. Keep all qualifications, roles, conditions, units, statuses and exclusions visible.
5. Test the rendering with a second-language reader and obtain qualified approval wherever conduct or decisions depend on it.

Scope, evidence and what the published standard does not promise

The official ISO 15704:2019 catalogue describes concepts and principles for enterprise architectures supporting development, integration, interoperability, human understanding and computer processing. It specifies requirements for models and languages expressing these architectures. It does not prescribe standard enterprises, standard organisational structures, standard processes, standard data or a particular enterprise modelling process. This distinction prevents overstating what the publication itself guarantees.

The specific reader question this article answers is translating enterprise architecture models, role labels, process maps and transition plans without confusing intended designs with deployed operations. It is written for business architects, technical writers, translators, ERP migration teams, consultants, operational managers and multinational programme teams. A reader should be able to use the worked examples to recognize what changes when a translation gets the scope, status or participant wrong. The examples are newly written training situations rather than extracts from paid normative clauses; they do not constitute certification, regulated technical advice, site permission or a replacement for primary evidence.

Chapter 1. First distinguish enterprise architecture from the particular enterprise

An enterprise architecture is a way of understanding and expressing an enterprise and its components, relationships and lifecycle concerns. A specific organisation then supplies its own people, divisions, operating rules and information systems. Translating the abstract reference as though it were a mandatory company structure assigns authority that is not in the source.

For example, a reference drawing showing ‘customer service function’ does not prove a company has a department with that formal name. A translated handbook must distinguish modelled capability from a named employee group.

Keep model type, intended audience, organisational context, and approval status in the heading or metadata so the reader can see whether the page describes an idea, a design or live reality.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 2. Translate mission, strategy and operational capability separately

A mission names the outcome the enterprise exists to achieve. Strategy concerns how it will pursue that outcome. Capability is an ability supplied through resources and practices. Process is an organised sequence of activities. These terms overlap in casual speech but are not interchangeable in a model.

A translation that substitutes ‘workflow’ for ‘capability’ can suggest a list of steps where the model intended an organisational ability. The wrong category then distorts investment decisions and responsibility maps.

Use a controlled concept list before choosing target-language words. Where no stable one-word equivalent exists, retain an original term plus a succinct definition.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 3. Preserve stakeholder concerns and system boundaries

A model’s boundary determines which organisation, partner, process, data flow and technology it represents. An external supplier can appear as an actor without becoming an internal department. Stakeholders can influence requirements without performing the modelled work.

A diagram showing a regulator’s review does not make the regulator a member of project management. A ‘customer’ entity may be an external party, not an employee. Translate each relationship according to its role in the particular model.

Review arrows, ports, data interfaces and legend definitions alongside visible text. These graphical features often carry more meaning than labels.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 4. Explain ownership, responsibility and authority with care

An accountable process owner, an individual who performs a task, an approver who authorises a decision, and a system that sends a message can all be different actors. Some languages use a convenient general noun equivalent to ‘responsible’ that conceals those differences.

If a bilingual contract replaces ‘approver’ with ‘operator’, a translated workflow could appear to let an operator release a transaction without the source approval. The change is significant even if surrounding sentences sound natural.

Use an approved responsibility matrix and keep authority verbs such as may, must, consults and approves visible. Do not infer accountability from a person’s job title alone.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 5. Keep as-is, to-be and transitional states in view

Architectural models frequently represent an existing state, an intended future state and a series of transitions. A target model can be approved but not yet built. A current-state process can be documented as flawed without being recommended. Wording must reveal this status.

A system in a planned future architecture should not be labelled ‘deployed’. A current bottleneck should not become a required control. Distinguish proposed, tested, approved, installed, active and retired statuses.

Ask for the effective date and decision record. Do not equate the date of an architecture slide with the date a capability entered production.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 6. Translate triggers, inputs, outputs and exceptions in process maps

A process description depends on the event that starts it, the information it consumes, the decision it makes and the output it produces. Translating an arrow as a simple sequence can hide a conditional branch or exception path.

A payment release after approval is not the same as payment release followed by approval. A ‘rejected’ request may return for correction rather than terminate permanently. These relationships have direct consequences for implementation.

For each flow, read the diagram as an executable story and test whether the translated captions support the same allowed paths.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 7. Separate human procedure from system interoperability

An enterprise may use people, machines and applications to carry out one business purpose. Interoperability is about how different components can work together; integration concerns the connections and coherence that enable it. Translating a technical API action as a human duty can create unsafe instructions.

When one application ‘publishes an event’ and another ‘consumes’ it, those words have technical meanings. A public-facing explanation may translate the concept into natural language but must keep the actual actor and payload relationship clear.

Prepare two layers if necessary: a readable explanation for managers and a controlled technical label for implementers. Ensure both refer to the same model element.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 8. Protect lifecycle, evidence and version history

Enterprise models are revised as systems and organisations change. A diagram can be historical evidence rather than the authoritative current process. A new version may supersede a design without changing contracts or live control arrangements.

An old procurement chart can be valuable as a record, but publishing it without a superseded label makes it look like current policy. The review should check source date, approving authority and deployment status independently.

Use version identifiers, change-request numbers and release boundaries in the translation package. Never silently remove exceptions, withdrawn claims or unresolved decisions.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 9. Check whether a model is explanatory, normative or illustrative

Architecture documents contain requirements, options, examples, assumptions, scenarios and aspirational statements. Each has a different level of obligation. A phrase that says a system ‘could’ implement a service cannot automatically become ‘must implement’.

Likewise, a hypothetical reference arrangement is not an enforceable rule merely because it appears inside a technical standard. The actual governing contract, policy and adopted architecture determine operational authority.

Review modality, source status and cross-references. Where a model is a teaching example, label it as an example and avoid claiming formal validation.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

Chapter 10. Design a multilingual model review that tests actual decisions

Some errors become visible only when a stakeholder uses the translation to assign a task, approve spending or build an integration. Re-reading for fluent English will miss many of these. Put a decision test next to each important translated relationship.

For a procurement flow, ask who may approve a purchase. For an IT transition, ask which system is live on the relevant date. For a service catalogue, ask who receives a failure alert. Answers should match the source model.

Have a subject specialist and language specialist jointly sign consequential roles and milestones, with unresolved differences documented instead of buried in smooth prose.

Make a small comparison chart for this chapter: source phrase, intended subject, target phrase, condition, responsible verifier and publication decision. Reading the source and target aloud can reveal clumsy wording, but cannot on its own verify responsibility, status or evidence. The crucial question is whether a reasonable reader could now perform an action, make a claim or assign authority that the original did not permit. If the answer might be yes, the translation needs further review even when its grammar is beautiful.

28 worked translation scenarios with editorial repairs

The scenarios below are fictional, independent teaching cases. Their source descriptions are not quotations from the cited ISO publication. They give a concise, reader-facing English repair, explain the tempting mistake and identify an actual verification task. Teams translating into French, Russian, Mandarin, Tamil, Arabic, Spanish or any other language should carry the same distinction into the target language rather than assume that English syntax is a universal template.

Case 1. Target model mistaken for a live process

Situation in the source. A slide calls the diagram ‘Future state, planned for 2028’. What goes wrong. Translating it as today’s operating process activates nonexistent roles.

Clearer translation or explanatory gloss. “’Proposed future operating model, targeted for 2028; not yet deployed.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Check the implementation milestone and approved scope before describing live responsibilities. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 2. The process owner label

Situation in the source. A chart names the ‘process owner’ above tasks performed by clerks. What goes wrong. Calling the owner the individual operator assigns every action to a supervisor.

Clearer translation or explanatory gloss. “’Process owner: accountable for governance; task performers shown separately.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Confirm the responsibility matrix and local governance definition. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 3. An external procurement supplier

Situation in the source. A supplier appears as a lane outside an enterprise boundary. What goes wrong. Translating the supplier as a company department creates false internal authority.

Clearer translation or explanatory gloss. “’External supplier sends confirmation to the enterprise procurement team.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Inspect swimlane boundaries and legal entity identifiers. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 4. A business capability

Situation in the source. A presentation says the company needs ‘customer insight capability’. What goes wrong. Rendering capability as a fixed software tool collapses people, practice and data into an app.

Clearer translation or explanatory gloss. “’Ability to analyse and act on customer insight; implementation to be determined.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Ask whether a product has actually been selected. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 5. A system integration point

Situation in the source. Diagram says an application publishes an order event to a message broker. What goes wrong. Saying a clerk sends the event turns machine behavior into a human task.

Clearer translation or explanatory gloss. “’The order application publishes an event for connected systems.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Check actor type and interface specification. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 6. Authority to approve

Situation in the source. Source states ‘reviewer recommends; director approves’. What goes wrong. Translating both verbs as ‘confirms’ obscures the authorizing decision.

Clearer translation or explanatory gloss. “’Reviewer recommends; director gives final approval.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Test the authority chain against the signed policy. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 7. A conditional branch

Situation in the source. The process moves to manual review if a verification score falls below a defined threshold. What goes wrong. Translating ‘if’ as ‘after’ makes every case go to manual review.

Clearer translation or explanatory gloss. “’Send for manual review only when the verification condition is met.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Verify threshold ownership and branch direction in the model. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 8. Exception handling

Situation in the source. ‘Rejected invoice returns to supplier for correction’. What goes wrong. Rendering the outcome as final cancellation destroys the rework path.

Clearer translation or explanatory gloss. “’Rejected invoice is returned for correction and may be resubmitted.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Check the process’s exception loop and restart conditions. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 9. A proposed shared service

Situation in the source. An enterprise plan says subsidiaries could centralise payroll. What goes wrong. ‘Subsidiaries must centralise payroll’ promotes an option into a mandate.

Clearer translation or explanatory gloss. “’Centralised payroll is a proposed option for the subsidiaries.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Inspect the decision record for approval. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 10. Legacy system status

Situation in the source. The diagram shows a CRM platform marked ‘retire after data migration’. What goes wrong. ‘Retired’ makes users abandon a still-live application.

Clearer translation or explanatory gloss. “’Scheduled for retirement after the verified data migration.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Confirm transition dependency and live service date. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 11. Two legal entities

Situation in the source. A holding company and operating company share a brand name. What goes wrong. Collapsing both into ‘the company’ misallocates contractual obligations.

Clearer translation or explanatory gloss. “’Holding entity’ and ‘operating entity’ with their legal names.” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Review registry identifiers and which entity signs each agreement. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 12. A role without a named person

Situation in the source. Source assigns ‘service manager’ as a modelled role. What goes wrong. Translating it as a specific employee’s title invents staffing.

Clearer translation or explanatory gloss. “’Service-manager role, holder not identified in this reference model.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Ask whether the role is organisational or instantiated. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 13. A planned audit control

Situation in the source. New operating design adds an audit checkpoint after transfer. What goes wrong. Describing it as an existing checkpoint suggests controls already work.

Clearer translation or explanatory gloss. “’Planned post-transfer audit checkpoint.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Compare the model against approved controls in production. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 14. A data definition

Situation in the source. In one system, ‘customer’ means bill-to account; in another, end user. What goes wrong. A single target term makes conflicting data domains appear identical.

Clearer translation or explanatory gloss. “’Billing customer account’ versus ‘end user’, per the respective data dictionaries.” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Trace identifiers through exchange mapping. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 15. A parent-child process

Situation in the source. A diagram decomposes ‘fulfil order’ into picking, packing and dispatch. What goes wrong. Translating components as three independent programmes loses their parent outcome.

Clearer translation or explanatory gloss. “’Subprocesses within order fulfilment: picking, packing and dispatch.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Inspect model hierarchy and trigger. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 16. A simulated test

Situation in the source. A prototype was demonstrated under synthetic traffic conditions. What goes wrong. ‘Operating successfully at production load’ turns simulation into evidence of deployment.

Clearer translation or explanatory gloss. “’Prototype demonstrated in a simulated load test.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Examine the test environment and limitations. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 17. A planned exception deadline

Situation in the source. Management proposes a 48-hour escalation limit, pending legal review. What goes wrong. ‘Escalate within 48 hours’ looks like a binding current rule.

Clearer translation or explanatory gloss. “’Proposed 48-hour escalation rule; awaiting approval.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Confirm final policy and clock-start definition. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 18. An ERP dependency

Situation in the source. The finance process requires an identity-service interface before activation. What goes wrong. Hiding the dependency makes the process seem independently ready.

Clearer translation or explanatory gloss. “’Activation depends on the identity-service interface being available.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Verify the release checklist and integration owner. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 19. A customer journey

Situation in the source. A team models an illustrative user journey from inquiry to aftercare. What goes wrong. Treating every modeled step as guaranteed service creates false commitments.

Clearer translation or explanatory gloss. “’Illustrative customer journey; actual service terms vary.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Review whether the diagram is descriptive or contractual. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 20. Enterprise versus system architecture

Situation in the source. A group diagram includes warehouse, staff, supplier and automation software. What goes wrong. Translating the whole view as ‘software architecture’ hides nontechnical components.

Clearer translation or explanatory gloss. “’Enterprise architecture view including operational and IT elements.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Confirm diagram scope and stakeholder audience. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 21. A regional permission

Situation in the source. Regional teams may adjust procedures within global policy boundaries. What goes wrong. ‘Regional teams set any policy’ expands delegated authority.

Clearer translation or explanatory gloss. “’Regional teams may adapt procedures subject to global policy.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Check governance delegation and exception rules. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 22. A withdrawn architectural option

Situation in the source. Version 3 rejects a central database option discussed in version 1. What goes wrong. A translated summary that retains the old recommendation misstates the current decision.

Clearer translation or explanatory gloss. “’Central database option was considered and withdrawn.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Match summary to the authoritative revision. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 23. Resource capacity statement

Situation in the source. A model says the team needs capacity for 1000 transactions a day. What goes wrong. ‘The team currently processes 1000’ confuses requirement with measured performance.

Clearer translation or explanatory gloss. “’Target capacity: 1000 daily transactions.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Identify whether the figure is a goal, test result or observed metric. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 24. A named value stream

Situation in the source. The term ‘order-to-cash’ covers a sequence across sales, fulfilment and finance. What goes wrong. Treating it as a bank transaction alone narrows the modeled business process.

Clearer translation or explanatory gloss. “’Order-to-cash business process, from order receipt through payment collection.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Verify scope and exclusions. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 25. An automated decision

Situation in the source. A rule engine recommends a credit limit; humans must still authorize it. What goes wrong. ‘The engine approves credit’ changes the responsible decision-maker.

Clearer translation or explanatory gloss. “’Rule engine proposes a credit limit for authorised human approval.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. A translation becomes misleading when it replaces a relationship with a more familiar word but leaves the reader with a different decision.

Specific quality-control test. Review decision rights and system action. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

For a practice review, ask two readers to paraphrase the original and the repaired version separately. Compare what each believes is established, and identify the exact word where an unsupported inference enters. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 26. A transition wave

Situation in the source. ‘Wave 2’ is a group of rollout sites, not the second week. What goes wrong. Translating wave as a calendar week creates schedule errors.

Clearer translation or explanatory gloss. “’Second rollout group (Wave 2), per deployment plan.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The difficulty here is the implied assertion. A sentence may be technically fluent while changing whether a fact is measured, permitted, planned or merely described.

Specific quality-control test. Inspect the programme’s wave definitions. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Cross-check the translated sentence with its linked table, diagram, map, permission record or technical file. A correct narrative beside an incorrect label is still a defective publication. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 27. Cross-functional reporting

Situation in the source. A security analyst reports operationally to the IT team and escalates incidents to risk management. What goes wrong. Rendering both paths as direct line management merges different relationships.

Clearer translation or explanatory gloss. “’Operational reporting to IT; incident escalation to risk management.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. The right improvement is often a small qualifier, a clearer actor or a concrete referent. Those few words keep the published claim within its source boundary.

Specific quality-control test. Verify arrow types and role legend. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Write an issue-log entry containing the source phrase, risk, approved rendering and open question. This makes the work reviewable when a later version changes the context. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

Case 28. A system boundary during outsourcing

Situation in the source. A call centre moves to a vendor but data ownership remains with the original company. What goes wrong. Treating operations transfer as data-ownership transfer changes governance.

Clearer translation or explanatory gloss. “’Operations outsourced; data ownership remains with the named controller.’” This version names the actor, state, evidence or boundary rather than polishing away the difficult distinction. An editor should ask what the reader could reasonably do next. If the target version invites an unauthorised or unsupported action, it has not preserved meaning.

Specific quality-control test. Confirm contract, data roles and transition date. The reviewer should compare the claim with the authoritative record, not only with the words on a screenshot. Record why the repaired language matters to this actual situation; a reader who has never seen the source must still be able to infer the correct limits.

Avoid solving the problem by adding a dramatic general warning. Preserve the specific limitation that matters to this case and use calm, precise language that readers can act on appropriately. Then try a second version for a nontechnical audience. It may use shorter words and a different sentence order, but it must preserve the same rights, conditions, scope, date and evidence status. If the brief leaves any of these unknown, the honest resolution is a query to the owner, not an invented detail.

How to plan the bilingual translation package

Start with a scope sheet rather than a term list. Name the source document, version, intended users, target locale, delivery channels and authority for corrections. Ask whether the text is a conceptual guide, a public notice, a contractual condition, a system specification or a historical record. A one-paragraph briefing can prevent a costly category error: a translator might otherwise write a proposed requirement as an approved instruction or a report of one measurement as a promise of universal performance. Include the source owner in the review at the beginning, not just at the last approval stage.

Create a source-to-target matrix with columns for original wording, concept definition, documented evidence, proposed gloss, audience interpretation and reviewer. Separate facts from editorial explanation. Where text contains named actors, model boundaries, units, thresholds, jurisdictional references or dates, assign those fields to a domain expert for verification. It is perfectly reasonable for the translation itself to be reader-friendly while the matrix remains technical. The matrix allows the editorial team to explain exactly why a seemingly fluent alternative could change the record’s meaning.

For recurring terminology, keep a termbase that stores definitions and examples of permitted use, not just dictionary replacements. Include forbidden false friends where one word in the target language collapses two separate functions. Preserve the original identifiers, document control labels and source citations. If a reviewer changes a term to improve accessibility, check every related caption, FAQ, table, image description and downloadable file for consistency. A multilingual glossary is valuable only when its entries remain anchored to real concepts.

Finally, test the published surface. Ask a target-language reader what the translated item tells them to do, whom to contact, what status to expect and what is still uncertain. In a business process, test role and action; in an environmental notice, test use and access instructions. Open mobile versions, printed signs, graphics and exported data. If two versions lead a user to different decisions, the release should be blocked until the owner settles the discrepancy. Keep approval records and a date for later review because live organisations and projects change.

Field workbook: five checks every translator should perform

1. Identity. Are the named organisations, model elements, places or assets the same as in the source?
2. Status. Does the text distinguish proposed, planned, approved, tested, active, suspended, retired or historical records?
3. Conditions. Are exceptions, limits, exclusions and qualifications still attached to the relevant claim?
4. Evidence. Can a second reviewer trace statements to an identified observation, requirement, model or approved instruction?
5. Action. Would a real reader make the same permitted decision after relying only on the target language?

Four-week learning programme

Week one: select several permitted public source records in this specialty. Write a concept map identifying actors, entities, conditions, evidence and document status. Draft a glossary of terms whose ordinary-language meanings might be misleading. Ask a subject specialist to mark any unsupported assumptions before translation begins.

Week two: prepare bilingual versions of at least twelve short but different statements. For each, write one deliberately overconfident translation and one evidence-faithful repair. Compare them by asking exactly which authority, scope or outcome changes. The exercise is complete only when a second reader can explain the difference without seeing your working notes.

Week three: move the approved statements into the surfaces where users actually encounter them: a diagram, sign, data table, public information page, short mobile description or operational briefing. Test headings, labels and explanatory notes independently. Check that visible text, screen-reader wording and any linked original record tell the same story.

Week four: run a blinded reader interpretation test, followed by a technical review. Give reviewers the target version without the source and ask what it means and what it authorizes. Reconcile their interpretation with the source and maintain a defect log. Revise the termbase, establish an owner for future corrections and publish only the version whose evidence and roles are traceable.

Frequently asked questions

Does ISO 15704 prescribe how every organisation should be structured?

No. ISO’s published scope explicitly excludes standard enterprises, standard organisational structures, standard enterprise processes and standard enterprise data. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

Are ‘current state’ and ‘target state’ interchangeable?

No. One describes a documented present situation and the other a desired future design. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

Can ‘responsible’, ‘accountable’ and ‘approver’ be translated with one word?

Only if the documented governance context truly makes them equivalent. In many workflows they denote separate decision roles. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

Should the enterprise model be treated as software architecture?

Not automatically. Enterprise architectures can include people, processes, organisational arrangements and systems. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

Can I change arrow directions when rearranging a translated diagram?

No. Direction, branching and actor relationships must preserve the source semantics. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

How do I treat a hypothetical future-state diagram?

Label it as proposed, preserve its dependencies and check whether it has approval for implementation. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

Who reviews this translation?

At minimum a qualified language reviewer and a model owner; consequential decisions may require policy, legal or engineering approval. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

What is the most damaging translation error?

Making an imagined design sound like a current requirement, or changing who is empowered to decide and act. For an actual high-stakes document, check the source’s governing version and seek the responsible specialist’s decision rather than generalise beyond the evidence described here.

Conclusion: translation is accountable to the source

A translated architecture should allow a reader to reconstruct the same organisation, process boundaries, possible future and authority relationships as the source, without mistaking a reference model for live operations.

The most practical lesson is that precision is not the same as rigid word-for-word literalism. A natural explanation can be accurate if it retains the actors, boundaries, exceptions and source status. Equally, technical-sounding words can be wrong if they shift responsibility or imply verification that never happened. A careful translation moves the same claim to the reader’s language while leaving its evidence and limits intact. When the source is unresolved, a transparent query and a qualified statement are better than an attractive invented certainty.

Primary reference and related learning resources

Official standard catalogue consulted on 5 November 2026: ISO 15704:2019 — public ISO scope and edition record. All worked scenarios in this article are original teaching examples, not quotations from protected ISO clauses. Project owners must apply the relevant licensed standard, applicable local regulations and qualified expert judgment for formal implementation or public safety decisions.

The parent for this specialist translation branch remains The Master Art of Translation (EDKSG-TRANS-MASTER-WORLD-000). The protected Vocabulary Learning Hub develops terminology and conceptual understanding. The existing How English Works: Translation and Meaning Across Languages explains the sentence-level transfer of meaning. This post addresses one document-and-decision search intent without replacing or restructuring those canonical learning resources.