How do you translate an accessible PDF into another language without breaking reading order, screen-reader pronunciation, alternative text, tables or interactive forms? A reliable PDF translation workflow treats visible sentences and the underlying tagged PDF structure as one multilingual publishing system. This guide explains how to translate a PDF for accessibility, how PDF/UA-2 relates to ISO 14289-2:2024 and PDF 2.0, how to preserve structural tags and /Lang metadata, and how to test a translated PDF before release.
The hidden problem is that a PDF can look perfectly translated while saying something different to a person using speech output, keyboard navigation, braille or text extraction. A two-column brochure can read across both columns, a translated diagram can have old-language alternative text, a table can lose its header associations, and a consent form can describe one option visually while exposing another name to assistive technology. A font that renders correctly is not proof that its Unicode text is extractable. Translation quality and accessible-document quality must therefore be judged together, but with separate evidence.
Quick method: inventory the source file and its editable source; identify the target language and audience; record the meaning and semantic purpose of headings, lists, figures, tables, links and fields; translate and revise the human-readable content; rebuild or preserve proper tags and reading order; update document and passage language metadata; localize alternative descriptions and control labels; test text extraction, navigation, focus and speech; check machine-detectable PDF/UA requirements using a suitable validator; perform human review against representative tasks; and retain a release log that identifies what was actually checked. This is a technical translation-production method, not a claim that every tagged file automatically satisfies every accessibility requirement.
Architecture: EDKSG-TRANS-MASTER-WORLD-1060. This standalone child addresses multilingual accessible PDF delivery. It does not replace the translation foundations at Master Art of Translation | The Complete System for Moving Meaning Between Languages, the protected vocabulary curriculum, or the protected English learning architecture.
Definition lock: accessible PDF translation, PDF/UA-2 and multilingual equivalence
In this article, accessible PDF translation means producing a target-language document that conveys the intended content and operations through visual reading and available assistive-technology pathways. Translating words is necessary but insufficient. A heading must remain a heading in the structure; a figure must retain an appropriate textual alternative when needed; a table must preserve the relationships that let someone find the right value; and the default language and changes of language must be identifiable for reading tools.
PDF/UA-2 is ISO 14289-2:2024, the accessible PDF conformance standard based on ISO 32000-2, the PDF 2.0 specification. It describes the way a conforming PDF uses PDF technology to support accessibility. PDF/UA-1, based on PDF 1.7, is a separate specification for older workflows. The existence of PDF/UA-2 does not mean every legacy PDF must simply have its version number changed. A genuine migration requires technical and human review of the exported content and its semantics.
WCAG and PDF/UA answer overlapping but different questions. WCAG expresses broader accessibility success criteria for content; PDF/UA governs how accessible features are represented in the PDF format. A file may pass a machine check and still have misleading alternative text or unintelligible prose. Conversely, a readable PDF may still lack required structural mechanisms for a claimed PDF/UA profile. Do not collapse a validator report, a content quality judgment and a legal compliance claim into the same sentence.
A tag is a semantic structure marker in the PDF, not merely a visual style. A reading order is the sequence that makes meaning when content is consumed linearly or through assistive technology; it can differ from the order someone sees by scanning a page. A language tag identifies the natural language of a document or passage. Alternative text is an authored representation of information from an image in a context, not an inventory of every pixel. Verification means gathering evidence for a specific acceptance criterion, with results reproducible by another reviewer.
Why this subject deserves a distinct translation system
A PDF translation project usually starts with a source document, a requested language, a deadline and perhaps a translation memory. Accessible delivery introduces another kind of source: the relationships between words and their roles. Those relationships often disappear in copy-and-paste, machine extraction, rasterization and careless export. The correct unit of work becomes an ordered, labeled, multilingual information object rather than an unstructured page of sentences.
Consider a two-page medication-information leaflet. Its visible Spanish headings might be accurate, but a scanner or export tool may put ‘warnings’ after a footer, split the dosage table into disconnected lines, or retain an English description of the warning icon. The final release could pass a casual visual glance while harming readers who must navigate by heading or hear instructions sequentially. This is a publishing failure even if every translated paragraph has good grammar.
The reader’s search question is usually practical: how to make a translated PDF accessible; how to set a PDF document language; how to translate PDF alt text; how to repair tagged PDF reading order; or how to validate a PDF/UA-2 file. This chapter keeps those natural questions attached to one production route, distinguishing what the translator can verify from what requires a PDF engineer, accessibility reviewer, legal adviser or subject specialist.
The eight-part preservation model
1. Intent
State what the source element is supposed to communicate, instruct, permit, warn or request; translate its function before deciding its surface wording. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
A visible approval button and its accessible name must authorize the same act, not merely contain plausible text. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
2. Content
Preserve the propositions, qualifiers, units, names, references, negations and intended audience of human-readable language. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
A translated negative instruction must not turn into a polite recommendation when the difference changes the action. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
3. Semantics
Retain structure roles: Title or heading, paragraph, list, list item, figure, table, header, note, reference and form field as appropriate. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
A large bold line is not automatically a heading, and a graphic of a table is not a data table. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
4. Sequence
Protect logical reading order, including columns, captions, footnotes, callouts, fields and links. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
A person navigating linearly must receive the cause before the conclusion and the question before its answer. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
5. Language metadata
Specify the main document language and meaningful changes of natural language at the relevant granularity. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
A multilingual quotation requires an appropriate language indication where supported; simply changing the file name is not enough. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
6. Interaction
Keep names, values, states, help text, focus order, validation feedback and submission consequences aligned. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
The translated visible label must not disagree with the name announced to a screen reader. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
7. Rendering
Preserve character identity, font coverage, glyph shaping, line breaking, pagination, numerical notation and extractability. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
A glyph that appears correct but copies as a replacement character fails useful text representation. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
8. Evidence
Separate automatic structural checks from task-based human review and record known limitations. In production, give this preservation obligation an identifiable owner and a practical acceptance test. The original file may expose the information visually, structurally or both, so the reviewer must know which representation supplies the evidence. Record where the verified information lives, which target-language version was inspected, and whether the check was automatic, manual or a combination.
‘Validator passed’ is not a substitute for reading the alt text in context and completing a form with a keyboard. The safest response to uncertainty is not to invent a reassuring status. Record the ambiguity, compare it with the approved source, decide who is qualified to resolve it and block release if the unresolved ambiguity could change a user’s decision. This turns an abstract quality principle into a traceable gate rather than a slogan.
A repeatable production method from source intake to release
Production gate 01 — Receive and classify
Identify whether the artifact is an editable source exported to PDF, a structurally tagged PDF, a flattened file, or a scan. Ask for the source files when available. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not promise preserved tags when the source has none; schedule reconstruction as a separate workstream. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A source inventory, list of target languages and known limitations. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 02 — Define target users
Specify language varieties, scripts, intended literacy level, document purpose and the assistive-technology tasks that matter. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not assume one language label or reading mode describes every intended reader. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A brief approved by the content owner and an accessibility reviewer. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 03 — Inventory semantics
Map headings, lists, diagrams, numerical tables, links, cross-references, forms, annotations and language changes before translation. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not let visual similarity replace a relationship inventory. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A numbered content-object ledger with stable source references. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 04 — Establish terminology
Choose consistent target forms for concepts, abbreviations, measurements, roles, field labels and legally significant verbs. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not let a termbase force the same word where grammatical or legal context requires a different expression. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A term decision record with forbidden false equivalents and examples. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 05 — Translate for action
Translate each message with attention to who acts, what condition applies, when the action happens and what it changes. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not soften warnings, manufacture consent or convert an example into an instruction. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A bilingual comparison set with resolved queries. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 06 — Rebuild accessible structure
Use an appropriate tagged-PDF-capable authoring or repair workflow, keeping headers, lists, relationships and alt text aligned. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not copy visible text onto a PDF image and call the result an accessible document. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A target file with inspected structure tree and ordered content. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 07 — Apply languages and scripts
Verify document-level language and passage changes, font coverage, shaping, right-to-left or mixed-direction behavior as relevant. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not treat rendered appearance as proof of reliable character mapping. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: Language and extraction checks with samples from all scripts. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 08 — Test interactive content
Walk every essential link, annotation and form field by keyboard and through supported accessibility interfaces. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not assume a field is usable merely because it can be clicked with a mouse. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: Field-by-field names, states and actions acceptance record. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 09 — Perform automated validation
Run a validator suitable for the claimed PDF profile; retain failures, caveats and tool/version details. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not describe an automatic pass as proof that all human judgments succeeded. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A saved rule report with remediations linked to defects. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 10 — Perform human verification
Review meaning, structure, speaking order, alt text, content language and actual user journeys with representative technology. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not certify a document solely from a screenshot, visual proofread or validator icon. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: Signed review sheet describing tasks, environment and findings. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Production gate 11 — Release with provenance
Freeze the approved file, revision, target locale, source version, reviewer roles and known exceptions. Treat this as a decision gate: name the owner, list the incoming evidence and state the exact output expected. Translation work often becomes unreliable at handoffs, when everyone believes someone else checked the structural side. A gate is complete only when its output is accessible to the following reviewer and versioned against the file that is actually going to be published.
A specific failure to prevent: Do not substitute a later edited file without rerunning affected checks. The release manager should ask what concrete reader task would fail if this precaution were ignored. A response such as ‘the page still looks fine’ is not evidence. When the source itself is wrong, open a source correction rather than quietly making a translation-only fix that future versions will overwrite.
Retained evidence: A release manifest and a documented route for corrections. Recheck this evidence after any export that can regenerate the PDF structure tree. The next gate must not infer success from a previous file’s result; references and file hashes should point to the approved target candidate, not to a source-language prototype.
Source research and standards: what is authoritative and what is not
The authoritative starting points are ISO 14289-2:2024, the PDF Association’s accessibility resources, the PDF Association’s standards overview, W3C’s PDF accessibility techniques, W3C PDF19 for language changes, and veraPDF’s validation documentation. These references establish a technical vocabulary. They do not certify a particular document merely because the document’s author cites them.
ISO 14289-2:2024 was published in March 2024 and concerns accessible PDF 2.0 files. ISO 5060:2024 offers general guidance for evaluating translation output, including human translation, post-edited machine translation and unedited machine translation. Evaluation of linguistic output and validation of the PDF format are two different inspection tracks. Both can be necessary for a trustworthy multilingual deliverable.
W3C PDF16 addresses default document language; PDF19 addresses language for a passage; PDF3 discusses reading and tab order; PDF1 addresses text alternatives; PDF6 addresses table markup; PDF9 addresses headings; PDF10, PDF12 and PDF23 concern labels, roles, values and interactive controls. W3C techniques are examples of ways to satisfy relevant criteria, not a claim that using a named tool guarantees conformance. Project requirements should name the target profile and any broader content obligations separately.
Worked diagnostic atlas: sixty-one real production patterns
The scenarios below are fictional but technically plausible. Every case uses the same audit questions: what does a person need to understand or do, what hidden representation could disagree with the visible page, where might translation introduce that disagreement, what repair preserves both meaning and access, and what evidence would support release? The differences between cases matter. A chart, a form, a quotation and a reading-order defect cannot be fixed reliably by one universal ‘tag everything’ command.
Diagnostic workshop 1: meaning and intended action
These investigations separate visual translation from executable meaning. Work from the reader’s task backward to the source evidence, not from a green compliance icon forward to an assumption of success. Each fictional case includes an error path, a mechanism, a proposed repair, a release test, and a transfer question. The goal is to make a repeatable professional habit, not to present one particular editor as a mandatory product.
Case 01 — Cover title and document identity
Reader task and source situation. Consider a public safety manual whose long English title has been replaced by a shorter German display title. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the identity of the current translated edition, audience and revision. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the file metadata still announces the earlier English edition and a revision date belonging to another version. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is document title metadata and the visible cover heading. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the identity of the current translated edition, audience and revision. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare title, catalog metadata, bookmarks and download link on the exact shipped file. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a renamed file can still misidentify itself in assistive technology and search. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. reuse the test in any multilingual document library. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 02 — Executive summary lead
Reader task and source situation. Consider an environmental report that begins with a conditional conclusion before a numbered list of exceptions. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the condition that limits the headline finding. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translator moves the exception below a decorative photo and the reading stream announces the conclusion alone. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is logical paragraph and list sequence. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the condition that limits the headline finding. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. linear read-through with headings hidden, followed by a bilingual proposition check. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a caveat can change the conclusion when its position changes. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to disclosures, briefs and public advisories. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 03 — Two-column brochure
Reader task and source situation. Consider a service guide containing two parallel columns with separate workflows. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is each complete workflow before the next begins. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when an export reads the first line of column A then the first line of column B, producing a false mixed procedure. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is logical reading order in the tag tree and related text objects. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve each complete workflow before the next begins. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. extract and read each column consecutively, then perform the instructions from the reading stream. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. visual layout can conceal an unusable listening sequence. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to newsletters, leaflets and annual reports. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 04 — Hierarchical headings
Reader task and source situation. Consider a technical booklet with title, chapter, section and subsection labels. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is navigable levels that correspond to the document’s argument. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when enlarged target-language headings are styled as ordinary paragraphs while subordinate headings are tagged above their parents. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is heading structure and levels rather than font size. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve navigable levels that correspond to the document’s argument. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. navigate by heading with a screen reader and compare the outline to the approved hierarchy. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. large typography does not establish structural heading status. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to manuals and teaching notes. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 05 — Numbered procedure
Reader task and source situation. Consider a ten-step assembly instruction whose step four has two dependent substeps. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the required order and dependency of the substeps. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when a layout reset restarts numbering and a translated substep is announced as a sibling instruction. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is list and list-item semantics, labels and bodies. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the required order and dependency of the substeps. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. reconstruct the sequence without looking at the page and execute a harmless sample task. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. the wrong list hierarchy can cause an action to occur too early. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to installation and laboratory protocols. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 06 — Nested legal clauses
Reader task and source situation. Consider a terms document with sections, lettered subclauses and cross-references. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which condition or exception qualifies which rule. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when an automated translation rebuilds letters as plain punctuation and changes a cross-reference after a reflow. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is list nesting, stable clause identifiers and anchor destinations. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which condition or exception qualifies which rule. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. follow every cited clause and inspect the associated list hierarchy. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. an intact-looking paragraph can point to the wrong exception. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to policies and contracts with legal review. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 07 — Footnote marker
Reader task and source situation. Consider a research article whose footnote qualifies an estimated percentage. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the relationship between the numerical claim and its qualification. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the superscript marker remains visible but the accessible content no longer connects it with its explanatory note. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is reference markers, footnote semantics and navigation. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the relationship between the numerical claim and its qualification. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. navigate from the claim to the note and return; check spoken order. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. unlinked notes can overstate confidence in quantitative findings. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to scientific publishing. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 08 — Endnote collection
Reader task and source situation. Consider a historical essay with many citations moved to the back. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the provenance of each claim and its correct source note. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translated endnote labels change but internal anchors refer to the original numbering. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is cross-reference targets and consistent labels. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the provenance of each claim and its correct source note. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. sample beginning, middle and end notes with link traversal and source comparison. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. numbering changes are not merely cosmetic. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to archives and scholarly reports. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 09 — Pull quote and sidebar
Reader task and source situation. Consider a magazine article where a sidebar summarizes but does not precede the main narrative. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is main-text continuity and the supplementary role of the sidebar. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the sidebar becomes the first spoken paragraph and looks like a fact established before its evidence. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is Aside or equivalent appropriate structure and ordered content. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve main-text continuity and the supplementary role of the sidebar. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. read without vision and ask whether the sidebar interrupts a dependent sentence. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. supplemental information is not a replacement for the argument. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to magazines and brochures. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 10 — Decorative separator
Reader task and source situation. Consider a promotional document using a repeated ornamental star between sections. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the pause or division without unwanted narration. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when each decorative image is given a verbose alt description in the translated file. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is artifact classification where decoration carries no information. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the pause or division without unwanted narration. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. count interruptions in speech output and check section transitions. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. describing decoration may burden comprehension rather than improve it. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to presentations and illustrated guides. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 11 — Information-bearing icon
Reader task and source situation. Consider an equipment guide using a red triangle to mean immediate danger. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the action and seriousness signaled by the icon in context. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the image receives a generic description such as ‘red shape’ after translation. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is meaningful figure alternative and nearby warning text. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the action and seriousness signaled by the icon in context. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. ask a reader of the text alternative what danger and action it conveys. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. shape description alone need not communicate warning meaning. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to safety and emergency documents. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 12 — Complex process diagram
Reader task and source situation. Consider a six-branch decision diagram in a claims guide. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is branch conditions, outcomes and return paths. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the new language fits in each box but its arrows are omitted from alternative text. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is diagram alternative plus detailed adjacent explanation when necessary. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve branch conditions, outcomes and return paths. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. answer sample branch questions without relying on visual arrows. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a list of node labels does not explain how the process works. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to service workflows and decision trees. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Diagnostic workshop 2: semantic structure and sequence
These investigations separate visual translation from executable meaning. Work from the reader’s task backward to the source evidence, not from a green compliance icon forward to an assumption of success. Each fictional case includes an error path, a mechanism, a proposed repair, a release test, and a transfer question. The goal is to make a repeatable professional habit, not to present one particular editor as a mandatory product.
Case 13 — Chart with uncertainty
Reader task and source situation. Consider a climate report chart with median, lower bound and upper bound. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is magnitude, trend, uncertainty and comparison that support the prose. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translation describes a single trend and drops the uncertainty band. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is chart alternative, data table or extended description. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve magnitude, trend, uncertainty and comparison that support the prose. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. give a nonvisual reviewer the exact comparison questions used by the editor. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a summary that erases uncertainty can misstate the evidence. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to financial and scientific charts. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 14 — Table of figures
Reader task and source situation. Consider a manual with a generated list of illustrated examples. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is navigation to the intended illustrations. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when figure numbers translate correctly but the internal destinations are broken. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is link structure, labels and targets. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve navigation to the intended illustrations. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. activate a sample of first, middle and last figure entries. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a table of figures is a navigation system, not a decorative page. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to textbooks and reports. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 15 — Financial balance table
Reader task and source situation. Consider a statement with years in columns and revenue, costs and totals in rows. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the association of each amount with its period and account. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when numbers look correct but column headers are flattened into plain text. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is table, header cells and scope or equivalent relations. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the association of each amount with its period and account. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. navigate by cell using headers and reconcile a selected row total. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a figure without its year is an incomplete financial fact. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to budgets and corporate reporting. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 16 — Matrix with multi-level headers
Reader task and source situation. Consider a comparison table grouped by product family and performance tier. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which group and subheader qualify a cell. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translation moves one group heading visually but not semantically. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is complex table associations and explicit header links. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which group and subheader qualify a cell. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. query several intersections including the outermost group boundaries. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. spatial proximity alone is unreliable when reading by row or cell. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to technical specifications and procurement. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 17 — Table spanning pages
Reader task and source situation. Consider a catalog with a price table continuing across three pages. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is one continuous data structure across the intended pages. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when repeated headers are treated as a new independent table and one product row is split. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is table continuity, meaningful header handling and page breaks. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve one continuous data structure across the intended pages. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. navigate the split row and confirm the same table context survives. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. page breaks should not silently turn one record into two. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to inventories and long price lists. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 18 — Chemical units in a table
Reader task and source situation. Consider a laboratory report where concentrations use mg/L and µg/L. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is correct numerical scale, unit and quantity association. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translated note changes the symbol micro to a similar-looking unsupported character. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is text mapping, fonts and table relationships. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve correct numerical scale, unit and quantity association. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy and paste units and inspect code points; have a subject reviewer check magnitude. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. glyph similarity does not make quantities equivalent. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to medicine and environmental measurements. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 19 — Figure caption order
Reader task and source situation. Consider a handbook with a photograph, descriptive caption and an explanatory paragraph. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is what evidence the image provides and which caption qualifies it. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the caption is pronounced after the following paragraph rather than with its photograph. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is figure-caption relation and reading sequence. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve what evidence the image provides and which caption qualifies it. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. listen from the paragraph before the photograph through the caption. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. displaced context can make a correct image appear to prove another claim. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to photographic documentation. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 20 — Image containing text
Reader task and source situation. Consider a schematic with labels baked into its pixels. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is every operational label and its relation to the image. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when only the nearby paragraph is translated while the diagram retains old-language instructions. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is translated graphic and an accessible text equivalent. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve every operational label and its relation to the image. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare label inventory against the revised diagram and screen-reader output. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a visual localization can leave critical instructions in the wrong language. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to equipment diagrams and infographics. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 21 — Screenshot with state
Reader task and source situation. Consider a software tutorial showing a disabled Save button and an error notice. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which interface state the screenshot demonstrates. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the alt text describes only the program’s name and not the state required for the lesson. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is purposeful alternative text or equivalent instructional text. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which interface state the screenshot demonstrates. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. ask a nonvisual learner what step is blocked and why. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. screenshots must be described for the task, not the pixels. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to product documentation. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 22 — Equation in a figure
Reader task and source situation. Consider a statistical lesson showing an equation and its variable definitions. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is operations, symbols, and variable relationships. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the equation is converted to a decorative image and its accessible explanation is lost. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is an accessible mathematical representation suited to the toolchain. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve operations, symbols, and variable relationships. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy the expression and have a specialist verify spoken and Braille pathways where relevant. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. visual mathematical resemblance is insufficient. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to STEM papers and textbooks. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 23 — Unicode minus and hyphen
Reader task and source situation. Consider a physics sheet with negative values and hyphenated compound names. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the sign of each quantity and the grammar of each compound. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when export replaces a mathematical minus with a hyphen or removes it on extraction. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is character mapping and accurate text content. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the sign of each quantity and the grammar of each compound. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy negative values and compare Unicode and calculation results. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a small character error can invert meaning. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to equations and engineering tables. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 24 — Embedded abbreviations
Reader task and source situation. Consider a medical leaflet using an unfamiliar acronym and a translated expansion. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the referent and how it is introduced to readers. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the acronym is left unchanged but the expanded form no longer matches it. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is accessible text and any supported expansion mechanisms. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the referent and how it is introduced to readers. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. read the first use aloud and ask whether both forms identify the same concept. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. consistency cannot be measured by identical letters alone. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to specialized domains with expert approval. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Diagnostic workshop 3: multilingual rendering and language metadata
These investigations separate visual translation from executable meaning. Work from the reader’s task backward to the source evidence, not from a green compliance icon forward to an assumption of success. Each fictional case includes an error path, a mechanism, a proposed repair, a release test, and a transfer question. The goal is to make a repeatable professional habit, not to present one particular editor as a mandatory product.
Case 25 — Quoted foreign-language phrase
Reader task and source situation. Consider a literary guide with an Italian phrase inside an English explanation. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the quotation’s exact wording and linguistic identity. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translated PDF declares only the primary language and speech pronounces the quotation incorrectly. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is language override at the relevant span or structure element. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the quotation’s exact wording and linguistic identity. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. inspect language entries and listen to the phrase using supported voices. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. translation does not erase the fact that an embedded quotation is another language. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to language textbooks and archives. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 26 — Bilingual section switch
Reader task and source situation. Consider a policy that presents French and Arabic versions on alternating pages. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which language governs each section and its reading direction. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the document-level language is used as if every page had the same language. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is language declarations per passage or section as supported. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which language governs each section and its reading direction. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. navigate alternating sections with supported voices and inspect actual language properties. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. mixed-language publishing requires granular metadata. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to multilingual public notices. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 27 — Right-to-left mixed numbers
Reader task and source situation. Consider an Arabic technical manual containing model numbers, dates and Latin units. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the precise order of identifiers and units in context. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when bidirectional reordering changes the apparent sequence of a serial number. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is logical character order, shaping and text extraction. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the precise order of identifiers and units in context. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy serials, dates and units; compare with verified source records. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. correct display is not guaranteed by flipping page alignment. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to international engineering catalogs. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 28 — CJK line breaking
Reader task and source situation. Consider a Japanese handbook with dense paragraphs and narrow labels. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is readable segmentation and intended compound meaning. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the export inserts forced line breaks that cut a compound in a misleading position. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is line-break rules and underlying text sequence. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve readable segmentation and intended compound meaning. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. review page proofs and extracted text with native-language specialists. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. line wrapping is a content quality issue when it damages reading. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to Japanese and Chinese publications. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 29 — Indic script rendering
Reader task and source situation. Consider a Hindi education handout containing conjunct consonants. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the intended words and pronounceable text stream. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the chosen font renders a conjunct as detached symbols and extraction misorders characters. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is font support, shaping and Unicode mapping. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the intended words and pronounceable text stream. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare visual proof, copied string and speech output on each affected page. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. font selection is part of translation delivery quality. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to multilingual scripts with complex shaping. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 30 — Thai text segmentation
Reader task and source situation. Consider a Thai-language form where words lack spaces between them. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the readable instruction and its association with a field. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the narrow field wraps at arbitrary glyph positions and obscures a required instruction. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is script-aware line breaking and field label handling. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the readable instruction and its association with a field. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. test text extraction and reading in the target environment. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a layout engine designed for spaced languages may fail silently. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to languages with different segmentation conventions. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 31 — Font subset and extraction
Reader task and source situation. Consider a multilingual annual report using a decorative typeface for headings. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is recoverable target-language Unicode characters. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the visible characters look correct but copy as question marks. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is font embedding and character-to-text mapping. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve recoverable target-language Unicode characters. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy a representative alphabet, accented text, punctuation and numerals. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. OCR should not be the workaround for a broken digital font mapping. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to all non-Latin and accented scripts. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 32 — Automatic hyphenation
Reader task and source situation. Consider a narrow-column translated report with long compound words. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is unbroken lexical meaning and searchable text. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when soft hyphens become hard characters during extraction and break search. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is line breaking, text extraction and actual text where appropriate. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve unbroken lexical meaning and searchable text. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy words across line wraps and run literal search for approved terms. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a translated term that cannot be searched is less usable. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to dictionaries and reference guides. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 33 — Hyperlink descriptive text
Reader task and source situation. Consider a report containing thirty links all shown as ‘click here’ after translation. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the purpose of each link in its local context. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when links announce identical names without destinations or purposes. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is link annotation, /Link structure and meaningful text. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the purpose of each link in its local context. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. open a links list and choose the correct reference without surrounding text. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a working URL does not make an understandable link. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to digital reports and resource lists. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 34 — Internal cross-reference
Reader task and source situation. Consider a safety instruction saying ‘see the emergency contacts on page 12’. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the correct destination of a time-sensitive reference. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translation adds pages but the cross-reference remains at the earlier location. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is internal anchor or accurate page reference. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the correct destination of a time-sensitive reference. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. activate the link and also follow any textual page number in print order. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. layout changes can invalidate instructions outside the paragraph itself. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to handbooks and emergency plans. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 35 — Bookmarks after reflow
Reader task and source situation. Consider a long accessible PDF whose translated headings wrap onto two lines. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is efficient navigation to the exact target heading. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when export preserves obsolete bookmark destinations from the source edition. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is bookmark labels, hierarchy and destinations. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve efficient navigation to the exact target heading. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. use bookmarks to navigate several levels including the final appendix. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. bookmarks must be rebuilt from the approved target structure. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to long reports and manuals. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 36 — Page labels versus page number
Reader task and source situation. Consider a book with Roman numerals for introductory pages and Arabic numerals later. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the reader’s ability to find a cited location. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the reader’s page labels and spoken page numbers diverge after front matter expands. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is page labels, printed numbers and navigation consistency. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the reader’s ability to find a cited location. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. look up the same citation from the table of contents and a reader command. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a numerical page reference is a navigational promise. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to books and academic publications. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Diagnostic workshop 4: interaction and feedback
These investigations separate visual translation from executable meaning. Work from the reader’s task backward to the source evidence, not from a green compliance icon forward to an assumption of success. Each fictional case includes an error path, a mechanism, a proposed repair, a release test, and a transfer question. The goal is to make a repeatable professional habit, not to present one particular editor as a mandatory product.
Case 37 — Headers and footers
Reader task and source situation. Consider a multi-chapter report with running heads and confidentiality footers. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is chapter orientation without constant disruptive repetition. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when all repeated elements are announced between sentences in the translated reading stream. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is artifact treatment and meaningful-content exceptions. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve chapter orientation without constant disruptive repetition. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. listen across several page boundaries and verify important footer content remains available. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. not every repeated mark is disposable and not every repeated mark belongs in main reading order. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to policies and annual reports. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 38 — Glossary two-column layout
Reader task and source situation. Consider a bilingual glossary with source term, preferred translation and explanation. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the exact pairings and their explanatory relationships. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when source and target terms are read in unrelated streams rather than paired records. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is table or definition-list semantics as appropriate. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the exact pairings and their explanatory relationships. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. query five entries using linear and cell navigation, not just visual scanning. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a glossary that scrambles pairs teaches the wrong vocabulary. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to terminology resources. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 39 — Consent checkbox
Reader task and source situation. Consider a bilingual application with ‘I consent’ and ‘I do not consent’ checkboxes. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the user’s choice and its legal or practical consequence. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the accessible name of the negative option is copied from the positive option. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is form field label, checked state and association. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the user’s choice and its legal or practical consequence. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. choose each control through keyboard and listen to name and state before submitting. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. an inaccessible label can manufacture apparent agreement. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to consent and preference forms with legal review. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 40 — Radio-button group
Reader task and source situation. Consider a scheduling form offering three mutually exclusive appointment windows. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which question the mutually exclusive options answer. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translated options are visible but their group prompt disappears from the accessibility tree. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is group label, option name and selection state. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which question the mutually exclusive options answer. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. enter the group from the previous control and inspect announced context. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. choice sets need the question as well as the choices. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to surveys and registration forms. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 41 — Required form field
Reader task and source situation. Consider a translated payment form requiring a tax identifier. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which input is compulsory and what format is accepted. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when an asterisk remains visible but no required state or accessible instruction is provided. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is form field description, required state and help text. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which input is compulsory and what format is accepted. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. skip the field and inspect the error, focus and spoken feedback. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. requiredness is an operational rule rather than a typographic symbol. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to transactional forms. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 42 — Localized date field
Reader task and source situation. Consider a multilingual registration document asking for a birth date. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the date value and required format. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when an example displays 04/05/2027 without disambiguating day and month. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is form instruction, field format and validation error. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the date value and required format. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. enter dates with day greater than twelve and test error messaging. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. ambiguous dates create mismatches even when all words are translated. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to applications and schedules. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 43 — Localized decimal and grouping
Reader task and source situation. Consider a financial declaration using 1,234.50 in one locale and 1.234,50 in another. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the numeric amount and field acceptance rules. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when a translated form displays the localized example but parses it using the old locale. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is visible example, parsed value and announced instructions. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the numeric amount and field acceptance rules. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. enter boundary amounts and verify submitted value against display. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. language localization must not silently change numbers. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to financial and scientific forms. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 44 — Form validation message
Reader task and source situation. Consider a document with an email field and a warning displayed only after submit. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is why the data was rejected and how to repair it. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translated error appears visually but focus remains elsewhere and the screen reader announces nothing. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is error text, focus management and field association. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve why the data was rejected and how to repair it. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. trigger every relevant invalid input and recover without using a pointer. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. an error nobody hears becomes an inaccessible dead end. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to interactive PDF and web forms. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 45 — Signature and initials
Reader task and source situation. Consider a contract with independent signer and witness fields. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the identity and responsibility of each person signing. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translated labels are swapped while the visual boxes remain in original order. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is field labels, instructions and signature locations. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the identity and responsibility of each person signing. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. navigate the signing flow and ask reviewers to state which role acts at each field. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. signature errors can invalidate intended workflow. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to contracts with professional legal review. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 46 — Annotations and comments
Reader task and source situation. Consider a collaborative report with review annotations intentionally included in the final PDF. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the status and purpose of each annotation. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when some annotations are copied untranslated while others describe obsolete wording. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is annotation type, text, language and navigation. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the status and purpose of each annotation. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. open the annotations pane, inspect origin and intent, and test focus order. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. editorial comments may not belong in a public release at all. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to consulted documents and annotated editions. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 47 — Redacted source document
Reader task and source situation. Consider a translated case file with sensitive names blacked out. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is irreversible removal of protected content from all representations. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translator reproduces black rectangles but underlying extractable text remains present. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is redaction of content streams, metadata and attachments. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve irreversible removal of protected content from all representations. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. extract all text, inspect hidden layers and attachments, and conduct authorized redaction QA. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. visual concealment is not reliable confidentiality protection. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to legal and privacy-sensitive records. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 48 — Scanned report and OCR
Reader task and source situation. Consider an archival report with pages delivered as photos. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is faithful text before any translation is approved. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when automatic recognition mistakes a ‘0’ for an ‘O’ in identifiers and misses a warning line. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is OCR confidence, image-to-text comparison and subsequent tagging. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve faithful text before any translation is approved. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. double-check sampled names, numbers, negations and page starts against scans. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. OCR introduces source errors that no target-language polishing can correct. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to document archives. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Diagnostic workshop 5: verification and release governance
These investigations separate visual translation from executable meaning. Work from the reader’s task backward to the source evidence, not from a green compliance icon forward to an assumption of success. Each fictional case includes an error path, a mechanism, a proposed repair, a release test, and a transfer question. The goal is to make a repeatable professional habit, not to present one particular editor as a mandatory product.
Case 49 — Photocopied tables
Reader task and source situation. Consider a scanned statistical appendix with faint column rules. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is numerical cell identity and correct table topology. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when OCR merges adjacent cells and suggests a false time-series trend. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is table reconstruction from source images and verified data. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve numerical cell identity and correct table topology. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare row and column totals and reperform a selected calculation. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. machine-recognized rows require human checking before translation. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to historic financial data. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 50 — Transliterated proper names
Reader task and source situation. Consider a biography with Arabic names presented in Latin transcription. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is a stable identification of people across sections. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translator changes spellings inconsistently between headings, bookmarks and citations. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is approved naming policy and consistent visible and navigational strings. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve a stable identification of people across sections. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. search for each approved form and compare all references. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. an inconsistent name breaks cross-reference and search. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to historical and multilingual indexes. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 51 — Product identifiers
Reader task and source situation. Consider an equipment catalog in which item numbers resemble ordinary words. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is exact identity of the item being ordered or maintained. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translation changes an alphanumeric part number while localizing the adjacent description. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is protected identifiers and their labels. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve exact identity of the item being ordered or maintained. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare exported identifiers against source inventory using exact-match checks. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. IDs are not ordinary prose and should be isolated from translation. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to catalogs and technical manuals. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 52 — Document version matrix
Reader task and source situation. Consider a procedure collection containing versions for several target languages. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is which content revision the reader is following. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the translated file inherits a source revision footer after an updated instruction. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is version labels and traceability across filename, cover and metadata. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve which content revision the reader is following. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. reconcile source commit, target revision and each footer on sampled pages. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a newer date without revised content is not an update. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to controlled documentation. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 53 — Embedded audio transcript
Reader task and source situation. Consider an accessible PDF containing linked audio with a text transcript. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is equivalent spoken and written information including pauses that distinguish speakers. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when audio narration is localized but the transcript still follows the old-language edition. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is accessible links, language labels and transcript association. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve equivalent spoken and written information including pauses that distinguish speakers. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare a sample audio segment with the corresponding target transcript. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. multimodal consistency is part of the release. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to multimedia guides. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 54 — Speaker turns in interview
Reader task and source situation. Consider an interview archive whose transcript shows interviewer and respondent turns. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is who said each sentence and when a question became an answer. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when translated speaker labels are visually aligned but not announced between text blocks. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is meaningful paragraph and speaker-label structure. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve who said each sentence and when a question became an answer. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. listen to several exchanges without looking at the page and attribute quotations. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. lost attribution can reverse an argument or imply consent. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to testimonies and oral histories. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 55 — Academic citation direction
Reader task and source situation. Consider a multilingual research bibliography containing Hebrew and Latin titles. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is author, year, title, locator and searchable source identity. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when mixed-direction punctuation and numbers reorder the source citation. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is logical text order and script-specific typography. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve author, year, title, locator and searchable source identity. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. copy citations into a plain-text field and search the resulting identifier. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. visual bibliography correctness is not enough if it cannot be extracted. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to journals and theses. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 56 — Map legend and key
Reader task and source situation. Consider a disaster-preparedness map with colors, symbols and evacuation directions. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is the route, limitation and category represented by each symbol. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the legend is translated but a path label and its alternative remain outdated. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is map alternative, textual route and legend associations. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve the route, limitation and category represented by each symbol. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. ask reviewers to locate a named destination using nonvisual instructions. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. map colors without accessible text are not a full translation. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to safety and wayfinding documents. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 57 — Timeline diagram
Reader task and source situation. Consider a historical report displaying events above and below a central line. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is event sequence, concurrency and source uncertainty. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when alternative text lists dates without preserving which events overlap. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is structured chronological description and cross references. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve event sequence, concurrency and source uncertainty. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. answer questions about before, after and simultaneous events using text alone. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a chronological list cannot always replace a parallel timeline. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to history and project planning. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 58 — Overlaid watermark
Reader task and source situation. Consider a confidential draft with a diagonal ‘DRAFT’ watermark. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is whether readers may rely on the content as final. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the target edition visually removes ‘DRAFT’ but retains draft status in metadata and references. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is document status in text, metadata and surrounding publication. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve whether readers may rely on the content as final. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. check that the released edition is consistently approved or explicitly marked provisional. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. a watermark can change reader reliance and trust. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to policies and regulated communications. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 59 — Accessible source versus repair
Reader task and source situation. Consider a translator receives both an editable document and an old tagged PDF. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is a maintainable workflow for future revisions. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the team repairs the PDF while later regenerating it from the unchanged source. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is authoring semantics first, PDF export and final verification second. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve a maintainable workflow for future revisions. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. make a controlled source change and verify the exported structure survives. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. one-off repair is fragile when regeneration overwrites it. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to recurring publications. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Case 60 — Web and PDF counterparts
Reader task and source situation. Consider a public information service offers an HTML page and a downloadable PDF. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is one approved message across alternate formats. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the PDF translation differs from the web page on eligibility dates and warnings. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is content synchronization and format-specific accessibility checks. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve one approved message across alternate formats. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. compare unique claims and revisions across both published variants. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. matching the design is not the same as matching the content. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to any multichannel publication. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Diagnostic workshop 6: meaning and intended action
These investigations separate visual translation from executable meaning. Work from the reader’s task backward to the source evidence, not from a green compliance icon forward to an assumption of success. Each fictional case includes an error path, a mechanism, a proposed repair, a release test, and a transfer question. The goal is to make a repeatable professional habit, not to present one particular editor as a mandatory product.
Case 61 — Offline distribution
Reader task and source situation. Consider a translated disaster guide shipped on USB and printed on paper. The translator should first ask what the reader must be able to decide, understand or carry out. In this case the invariant is essential meaning without an internet connection. That invariant needs to hold in the visible target-language rendering, in a linear reading stream and in any applicable interactive control. A bilingual reviewer can judge the proposition; an accessibility reviewer can judge the manner in which the proposition becomes available.
Hidden failure. A plausible release error occurs when the PDF depends on a remote font or link for the only explanation of an emergency icon. This is not simply a spelling defect. It may change the identity of the information, the timing of a decision, the available choices or the reader’s route to an explanation. If the translated page is checked only visually, the problem can remain invisible to the team. Conversely, a structure validator can flag some mechanical conditions but cannot decide whether a translated statement actually carries the source writer’s intention.
Mechanism and boundary. The representation requiring explicit attention is embedded resources and accessible local alternatives. Ask whether the target file includes a reliable relationship between the displayed content and its underlying semantics. Treat text encoding, names, focus order, reading order and annotations as distinct properties even when they happen to be built by the same export tool. A convenient implementation shortcut is acceptable only if the resulting file passes the purpose-based check and still behaves consistently across intended readers.
Repair and controlled revision. Return to the approved source, isolate the faulty element, and write down the minimum correction needed to preserve essential meaning without an internet connection. Update the editable source when a maintained source is available; otherwise use a suitable accessible-PDF repair method and document the limitation. Regenerate and retest the target output rather than pasting over its appearance. If the issue involves a specialist claim, obtain its decision from the relevant content owner. Never silently transform an ambiguous source into a confident translation.
Practical acceptance test. open offline and complete the core reader tasks with network disabled. This check should be repeated on the deliverable rather than an earlier proof. For a meaningful sample, include first and last occurrences, long target strings, unusual scripts, exceptions, and cases where appearance and structure are likely to diverge. Save enough evidence for another reviewer to reproduce the finding. A machine pass, where applicable, is one column of that evidence; a task-based human pass is another.
Why it matters. accessibility should survive the intended distribution conditions. A defect is severe when it changes a consequential reader action, even if it affects only one character or control. Classify it by consequence and affected representation, not by the amount of visible ink. When a failure is discovered, check neighboring instances made by the same template, converter or translation-memory segment, because identical production assumptions can replicate the error across every language version.
Transfer and diagnostic question. apply to field manuals and resource-limited settings. Ask a reviewer who has not seen the visual source to explain the reader’s next step, then compare that explanation with the approved source and the rendered page. If the three disagree, the case remains open. If they align, retain the result and note what environmental assumptions were tested, including the reader technology and file version; that evidence is what makes the correction reusable.
Failure mode taxonomy: diagnostic symptoms and responsible routes
Valid tags, wrong meaning
Symptom: An alt-text string is present and syntactically accepted, but describes a harmless object where the picture warns of danger. Diagnostic route: Send the semantic decision to the content and language reviewers; fix the wording, then rerun structure checks and user tasks. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Correct prose, wrong order
Symptom: All text translated accurately, but the reading stream crosses columns or skips a conditional sentence. Diagnostic route: Send structure and sequence to the PDF authoring specialist; do not demand the translator solve tree structure without access. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Readable pixels, wrong characters
Symptom: Text displays correctly but copies with corrupted letters, missing signs or reordered numerals. Diagnostic route: Inspect font mappings and Unicode representation; test extraction before revisiting translation. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Accessible name disagrees with visible label
Symptom: A form field announces a different question from the one printed next to it. Diagnostic route: Fix the actual form field semantics and record the consent or transaction risk. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Automated pass, human failure
Symptom: Validation reports no machine-detectable violation but the chart alternative omits the comparison a reader needs. Diagnostic route: Retain validator results; require a human factual check against the intended reader question. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Human-readable, missing programmatic data
Symptom: A sighted reviewer can interpret colors or columns but an interface cannot expose their relationships. Diagnostic route: Reconstruct semantic relationships and retest the linear or table navigation task. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Wrong language inherited
Symptom: A paragraph in another language is pronounced using default-language rules. Diagnostic route: Apply justified document or passage language metadata and test with capable speech resources. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
OCR error imported as translation
Symptom: A recognition pass substitutes characters in a number or misses a negative word. Diagnostic route: Verify source transcription first; restore source fidelity before translating again. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Source corrected, target stale
Symptom: The master text changed after translation, but the target PDF has not been regenerated. Diagnostic route: Use controlled source-to-target version linking; block release of unmatched editions. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Old footer in new edition
Symptom: Metadata, headers, or cover still name a prior language or revision. Diagnostic route: Check all representations of title, version and status on the final bytes. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Template repaired only once
Symptom: A manual adjustment fixes one PDF but the next export repeats the defect. Diagnostic route: Repair the authoring template, document the converter settings and keep a regression test. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
Unsupported conformance claim
Symptom: A footer says ‘fully accessible’ despite only running an automatic check. Diagnostic route: Describe actual checks and limitations; never extrapolate to wider legal or format compliance. The quality record should identify which representation was observed, the user task affected, the correct expected behavior and the person authorized to accept or reject the repair. That distinction prevents language reviewers from being asked to certify code and developers from being asked to certify meaning. The next test should be a narrow reproduction followed by a cross-file regression check.
A useful reviewer question is: ‘What could a user reasonably do differently because of this error?’ If the answer changes a choice, a safety precaution, a legal consent statement, a quantity or a route to essential information, classify the matter as significant until a qualified review resolves it. When the answer concerns only style, still record the issue but avoid inflating it into a safety claim. Severity is tied to evidence and consequence, not to how alarming the defect report sounds.
How to build an evaluation specification that is honest about evidence
Translation evaluation and file-format validation must have independent acceptance criteria. A linguistic sampling plan can review meaning, terminology, fluency, style, consistency, omissions and additions; PDF-oriented testing can examine structure, links, language settings, text extraction, navigation and forms. They inform one release decision, but a pass in one track must not hide a failure in the other. An evaluator should be able to say ‘the translation is accurate but the reading order is wrong’ or ‘the file is mechanically valid but the warning text is misleading’ without contradiction.
Risk-based review begins by listing reader tasks. An educational worksheet must support reading questions and entering answers; a safety procedure must preserve sequence and conditions; an invoice must correctly associate amount, currency, item and payee; a legal document must accurately identify clauses and consent; and a research appendix must keep evidence traceable to its limitations. A uniform percentage sample can miss the only dangerous page. Supplement sampling with complete review of high-consequence elements and every interaction type.
For each identified defect, record an ID, source and target files, page or object, expected meaning, actual result, affected user route, severity rationale, proposed fix, owner, verification method, tested environment and disposition. A correction is not closed by changing a translation-memory entry. It closes when the revised target file demonstrates the expected behavior and the release manifest points to the new bytes.
Release checklist: the editor can use this without specialized terminology
- Does the target-language title match the delivered edition and document metadata?
- Can a reader navigate the hierarchy of headings without scanning every page visually?
- Do lists, multi-column paragraphs, tables, figures and notes read in a meaningful order?
- Are meaningful images and non-text visuals described for their informational purpose?
- Are decorative elements prevented from overwhelming speech output?
- Does the document have a correct primary natural-language setting, with meaningful language changes handled?
- Do all scripts render, copy, search and speak as expected in intended environments?
- Are field names, states, requiredness, errors and submit consequences accessible and translated accurately?
- Do bookmarks, link labels, targets and page references lead where they promise?
- Are calculations, units, identifiers and revisions consistent with approved sources?
- Has a validator appropriate to the claimed PDF profile been run and its result retained?
- Has a human reviewed meaning, alt text, reading order and representative reader tasks?
- Are unresolved defects recorded, assigned, and explicitly excluded from unsupported conformance claims?
- Is the exact final file, not only a staging proof, the version that was tested?
Twenty-five questions that translators and document teams actually ask
Question 1: Can I translate an existing PDF without the original source?
Sometimes, but treat it as a reconstruction project rather than a guaranteed one-click conversion. Inspect whether the PDF has reliable text and tags. An editable source normally gives a better place to repair headings, lists and language before re-export. A scan needs transcription or OCR verification before translation. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 2: Is PDF/UA-2 the same as WCAG 2.2?
No. PDF/UA-2 governs the use of PDF technology for accessible representation; WCAG expresses broader success criteria. Their concerns overlap, but the two labels are not interchangeable. State the specific standard, profile and assessment that your deliverable meets. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 3: Does a green validator report prove the PDF is accessible?
No. Automated checks are valuable for machine-detectable rules. They cannot by themselves judge whether an image description conveys the correct point or whether a translated instruction causes the intended decision. Combine automated results with human, task-based inspection. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 4: Must every picture receive alternative text?
Not every decorative image needs a spoken description. The question is whether the image carries information or performs a function. Information-bearing graphics require a meaningful alternative or accessible equivalent. Decorative repetition can instead be handled as an artifact when appropriate. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 5: Where does the document’s default language belong?
PDF provides document-level natural-language metadata, including /Lang. The correct value should represent the actual dominant language of that document. Important language changes within content need an appropriate representation where supported and required by the project. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 6: What if the document contains many languages?
Choose a defensible default language and represent significant changes at relevant spans or structure elements. Test with compatible assistive technologies and language resources. A multilingual title or quoted phrase should not be silently announced under the wrong pronunciation rules. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 7: Is the visual reading order always the spoken reading order?
No. Visual arrangements support scanning, whereas a linear reading route must make dependencies explicit. Multiple columns, sidebars, captions, notes and field prompts are common failure points. Test by listening or extracting order, not only by inspecting pages. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 8: How do I handle tables after translating the headers?
Check that data cells remain associated with their appropriate row and column headers. Multi-level and multi-page tables may need explicit structural decisions. A visually intact table can still be unusable for readers navigating cell by cell. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 9: Can machine translation be used on accessible PDFs?
It may assist the language stage when the project permits it, but it does not establish factual accuracy or preserve all PDF semantics automatically. Protect IDs, numbers and structures; revise the output; then inspect the newly exported document through separate linguistic and accessibility checks. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 10: What happens when the target text becomes longer?
Reflow may change pagination, captions, footnotes, bookmarks and reading order. Do not compress text until it is hard to read merely to maintain the source page count. Rebuild navigation and test the resulting layout and structure. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 11: What if the font supports the characters visually but copy-and-paste is broken?
Treat it as a representation defect. Inspect font embedding and character mapping; try extraction and search tests across each script. Use an authoring or export route that preserves correct Unicode rather than relying on an image appearance alone. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 12: Should I translate product codes and model numbers?
Usually protect identifiers as identity fields, subject to the product owner’s documented naming rules. Translate their descriptive labels, not the alphanumeric identity itself. A changed model number can produce an ordering or safety mistake. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 13: Is a PDF with OCR automatically accessible?
No. OCR may supply a text layer but cannot by itself prove correct wording, semantic order, table structure or language metadata. Compare recognized text against images, especially for numbers, negation and identifiers, then construct accessible semantics. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 14: What is the best test for a bilingual form?
Walk the form as a user would. Hear every control’s name and required state, enter data, trigger relevant errors and submit a test response. Then confirm that visible text, announced instructions and accepted values represent the same choice. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 15: Can I claim PDF/UA-2 after upgrading the file header to PDF 2.0?
No. Conformance is not a version-label edit. A target document must meet the applicable format and accessibility requirements. Evaluate its structure and behavior with the correct specification and suitable verification. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 16: Should I remove all running headers and footers from the reading stream?
Not indiscriminately. Repeated purely decorative content may be treated differently from essential information such as document status. Determine the informational purpose before making it an artifact or part of the main reading sequence. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 17: What if the target language is right-to-left?
Inspect logical character order, mixed-script identifiers, punctuation, dates, shaping and navigation as separate problems. Mirroring the layout alone cannot guarantee equivalent reading or extraction. Check target-language samples with competent readers. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 18: Why do bookmarks matter if headings are already tagged?
Bookmarks offer another navigation route in many PDF readers. Both can be valuable and can fail independently. A translated edition should offer consistent labels and destinations, especially when reflow changes pages. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 19: How can I check complex charts without writing paragraphs of alt text?
Define the question the chart answers. Use a concise relevant summary and, when needed, a structured data table or extended description. The alternative should preserve important comparisons, trends and uncertainty rather than enumerate every pixel. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 20: Do I need a human reviewer after automated testing?
Yes when claiming a meaningful accessibility and translation release, especially for semantics and user tasks that machines cannot judge fully. Choose reviewers with relevant language, domain and accessibility competencies. Record what they did rather than a vague assertion of ‘checked’. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 21: What about accessibility of links to external resources?
Make link purposes understandable in context and check that targets function. Where an external resource is essential, consider whether readers have a usable alternative, but do not misrepresent an external site as controlled by the PDF publisher. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 22: How should I score errors?
Use a documented severity scheme tied to reader consequence, purpose and affected representation. ISO 5060:2024 can inform human evaluation of translation output; it is not a PDF structural validation standard. Keep format failures and linguistic errors separately visible. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 23: How do I handle a source document that is already inaccessible?
Record the inherited defect and obtain agreement to repair the source where possible. A target translation should not be forced to replicate an avoidable access barrier. Distinguish source defects, translation errors and production errors in the audit trail. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 24: Should the translation memory store alt text?
It can store approved alternatives with strong contextual metadata, but the same image can require different descriptions in different uses. Review each reused segment for purpose, surrounding copy and target audience before approving it. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
Question 25: What does a realistic sign-off look like?
A release record names the exact final file, target language, source revision, linguistic reviewer, accessibility checks, validator version, tested reader tasks, known limitations and approval date. It does not promise universal access under every possible combination of software and technology. The operational question is which reader task would reveal a wrong answer. Document a reproducible check on the target file, specify the person who owns the decision, and retain the result for the next revision. That way the answer is not merely advice for one document: it becomes a local production rule that can be applied in future languages without silently reducing the quality threshold.
A four-week implementation and training plan
Week 1: discover the document’s true structure
Inventory a representative set of source files and extract their heading, list, figure, table, form and language information. Separate digital-born files from scans. Create an issue log. Select two or three essential reader journeys, and ask what a successful nonvisual reading experience would sound like. The deliverable is a baseline, not a conformance certificate.
For each training session, assign a meaning reviewer, a structural reviewer and a release owner, even if one individual has more than one role. Ask the team to demonstrate one actual reader journey and one failure that a visual scan missed. The learning target is not memorizing a software menu. It is the ability to explain how a translated proposition reaches a real reader, diagnose where that chain breaks, and verify the repaired file under a stated set of assumptions.
Week 2: design the bilingual production route
Create a short translation brief and decision log for terminology, protected IDs, locale rules, script support and potential content changes. Translate a controlled sample containing a long heading, a table, a figure and one interactive control. Make the authoring and accessibility teams exchange evidence rather than PDFs alone. The deliverable is a usable pilot with known risks.
For each training session, assign a meaning reviewer, a structural reviewer and a release owner, even if one individual has more than one role. Ask the team to demonstrate one actual reader journey and one failure that a visual scan missed. The learning target is not memorizing a software menu. It is the ability to explain how a translated proposition reaches a real reader, diagnose where that chain breaks, and verify the repaired file under a stated set of assumptions.
Week 3: inspect the complete target and repair patterns
Generate the full translated PDF using the approved workflow. Test linear reading, headings, language metadata, figure alternatives, table associations, copy-and-paste, links, and form behavior. Run machine checks appropriate to any claimed PDF profile. When a repeated defect appears, repair the upstream template rather than patching dozens of identical pages by hand.
For each training session, assign a meaning reviewer, a structural reviewer and a release owner, even if one individual has more than one role. Ask the team to demonstrate one actual reader journey and one failure that a visual scan missed. The learning target is not memorizing a software menu. It is the ability to explain how a translated proposition reaches a real reader, diagnose where that chain breaks, and verify the repaired file under a stated set of assumptions.
Week 4: release and make it repeatable
Review high-consequence content with a qualified subject owner. Retest corrected files, document any exclusions, freeze the final bytes and record the validation environment. Publish only a description of conformance that the evidence supports. Prepare regression examples for the next target language or edition so the process improves instead of starting from zero.
For each training session, assign a meaning reviewer, a structural reviewer and a release owner, even if one individual has more than one role. Ask the team to demonstrate one actual reader journey and one failure that a visual scan missed. The learning target is not memorizing a software menu. It is the ability to explain how a translated proposition reaches a real reader, diagnose where that chain breaks, and verify the repaired file under a stated set of assumptions.
A practical record template for real projects
Document identifier: [stable ID] Source revision and source language: [version / language tag] Target edition and target language: [version / language tag] Target PDF profile: [PDF/UA-1, PDF/UA-2 or no formal claim] Key audience tasks: [navigate / read / calculate / complete / sign] Semantic objects checked: [headings, lists, figures, tables, fields, links] Translation evaluator and method: [name / sample or full review / findings] Structural validator and version: [software / profile / rule report] Human accessibility review: [technology / settings / tasks / observations] Significant defects and dispositions: [ID / risk / owner / verified fix] Final file identifier: [hash or controlled release ID] Approver and date: [name / ISO date] Limitations or exceptions: [precise statement]
This template deliberately separates claims, observed evidence and responsibilities. Do not fill a field with ‘passed’ unless there is a specific underlying test and a corresponding output. For a PDF that contains personal or confidential information, retain review evidence securely and avoid recording more private content than the quality process needs.
Glossary of terms used in accessible PDF translation
Accessible name
The name exposed to assistive technology for a control or other interface item; it must align with what the control does. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Alternative text
Textual information serving the purpose of relevant non-text content; its adequacy is contextual. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Artifact
Content classified as not part of the meaningful reading structure, often decorative or redundant material. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Assistive technology
Software or hardware that helps a person access or operate content, such as screen-reading and Braille systems. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
BCP 47
A widely used language-tagging framework for identifying languages and regional or script variants. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Bookmark
A navigation entry linked to a document destination, often organized hierarchically. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Character mapping
The relationship between encoded PDF text and the text characters users extract, search and hear. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Document language
The default natural-language metadata that informs suitable presentation of a document. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Form field
An interactive control with name, state, value and often instructions or validation feedback. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Heading hierarchy
The semantic levels indicating section nesting rather than merely visual font sizing. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
ISO 14289-2
The 2024 international standard defining PDF/UA-2 accessible PDF requirements using PDF 2.0. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
ISO 32000-2
The international PDF 2.0 specification underlying the PDF/UA-2 profile. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
ISO 5060
General guidance for human evaluation of translation output, distinct from format validation. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Language of parts
The identification of a phrase or passage’s natural language when it differs from the surrounding content. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Logical reading order
An ordered presentation of content that preserves meaningful relationships when consumed sequentially. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
OCR
Optical character recognition, which estimates text from visual images and requires verification. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
PDF/UA
The PDF standards family addressing accessible, interoperable use of PDF features. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Reading stream
The sequence of information encountered in reading or accessibility extraction, not merely the visual page. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Role map
A mapping that helps interpret custom structure types against standard PDF structure semantics where applicable. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Screen reader
Software that renders interface information through speech or another accessible output path. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Semantic tag
A structure element identifying a content role, such as paragraph, list, table or figure. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Structural validation
Checking applicable machine-testable file and tag rules against a defined PDF profile. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Tagged PDF
A PDF representing logical structure and content semantics through a structure tree and related mechanisms. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Text extraction
Retrieving the encoded text and order from a document, which can reveal errors hidden by rendering. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Verification record
Traceable evidence connecting a test, file version, observer, expected result and finding. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
WCAG
Web Content Accessibility Guidelines, broader content accessibility criteria distinct from PDF/UA format requirements. In translation production, use this term only with the specific evidence or object it describes. Do not let a technical label stand in for the reader’s outcome: what matters is whether a person can obtain, understand and use the target-language information without a barrier introduced by the document’s translation or export.
Sources and careful scope of authority
- ISO 14289-2:2024 (PDF/UA-2) — official ISO standard page
- PDF Association — accessible PDF standards, tagged PDF and working-group guidance
- PDF Association — standards bundle and historical relationship between PDF/UA-1 and PDF/UA-2
- PDF Association — Tagged PDF Best Practice Guide
- W3C WAI — PDF technique PDF3, reading and tab order
- W3C WAI — PDF technique PDF19, language of passages
- W3C WAI — WCAG PDF techniques collection
- veraPDF — machine-testable PDF/UA-1 and PDF/UA-2 validation guidance
- ISO 5060:2024 — evaluation of translation output
The named specifications and technique documents are sources of terminology and scope, not evidence that this page or any particular downstream file has been independently certified. Product support and validation capabilities may change; evaluate the documented features of the actual tool version used. Historical examples in this guide are invented for teaching and must not be treated as real incident reports.
Next routes: what to learn after accessible PDF translation
For the underlying logic of moving intent, structure, pragmatic meaning and evidence between languages, return to Master Art of Translation | The Complete System for Moving Meaning Between Languages. The root is the parent architecture; this PDF/UA-2 chapter contributes one specialized, independently usable workflow and does not rewrite that parent.
When a translation error begins as misunderstanding of a term, idiom, register or semantic contrast, continue to the protected Vocabulary Learning Hub. That route owns vocabulary acquisition and retrieval. If the problem is sentence meaning, register, grammar, reading comprehension or English-language explanation, the protected How English Works | Translation and Meaning Across Languages route supplies relevant language foundations. These are outward pointers only: no content, title or ownership in those existing learning worlds is modified by this article.
The final rule is simple to apply and demanding to prove: a translated PDF should give the target reader the same essential information, decision paths and controls through the reading methods the document promises to support. Preserve meaning first, represent it structurally, inspect the real reading experience, and publish only the claims that the final file’s evidence justifies.