METS XML structural maps for digital archives: this guide answers how to translate METS structMap labels, file pointers and archival navigation correctly. The safest method is to localize digital-library page titles, section labels and access notes while preserving the METS structural map, hierarchy, file references and sequence. It is written for a digitisation librarian, collections translator, archival engineer or metadata editor, with worked examples and a practical translation quality assurance checklist, not a substitute for a validator or a system owner.
The visible problem is that a harmless-looking translation can sever a page link, reorder chapters, confuse physical pages with logical sections or hide the archival master behind a thumbnail. A translator may understand each ordinary word yet miss the technical relationship between a label, its code and the downstream action. The solution is to separate human explanation from machine identity, preserve what the data actually asserts, and check the result in the document’s real workflow.
For accurate technical translation and localization, begin with METS XML element names, ID and IDREF links, FILEID targets, TYPE and ORDER semantics, file-group associations and object provenance. Then identify which natural-language phrases readers genuinely need in their own language. This page explains the first principles, a repeatable twelve-stage method, field-level examples, common errors, practice activities and next reading routes. It deliberately keeps this subject separate from generic advice about translation style, idioms or glossaries.
Quick answer: preserve stable identifiers; translate explanatory wording only when its function is known; verify every important relationship against the primary source; record unresolved ambiguity instead of inventing confidence; and test the final edition with a domain reader. A translated document succeeds when someone can make the same justified decision from it as they could from the original.
- Canonical job: localize digital-library page titles, section labels and access notes while preserving the METS structural map, hierarchy, file references and sequence.
- Do not alter: METS XML element names, ID and IDREF links, FILEID targets, TYPE and ORDER semantics, file-group associations and object provenance.
- Best first action: copy the source, identify the version and separate labels from identifiers.
- Best final action: structural checks, semantic checks and a documented handoff.
The essential distinction: translation is not protocol conversion
Natural language is flexible because speakers can express one purpose in several ways. Structured records are constrained because downstream systems may interpret particular tokens, offsets or relationships literally. The human text around a record may be translated freely within the boundaries of meaning, but its functional identifiers are not prose. This is why the familiar instruction to “translate the meaning rather than the words” requires an extra safeguard in a machine-readable environment: determine exactly where the meaning lives and where the executable structure begins.
A correct target-language phrase is not proof of a correct operational result. Consider a label that appears near a numeric value. Before choosing a synonym, establish whether the number is a count, identifier, date, position, measurement, quality flag or ordering key. If it controls a dependency, a single misplaced relationship can cause a larger mistake than a slightly awkward adjective. An audit must therefore ask two questions: does the reader understand the message, and does the record still mean the same thing to the consuming system?
Document authority also matters. A public specification can define the general model, while an organisation’s implementation profile determines supported values, optional fields or local naming conventions. Translate neither a missing value nor a guessed default into an affirmative fact. If the profile is absent, mark the decision as provisional and identify the field whose interpretation needs confirmation. This is a positive professional outcome: a transparent uncertainty is safer than a polished falsehood.
First principles: what stays fixed and what can change
1. Identity: a principle that can be tested
A string used as an ID, address, reference, or machine key is a locator, not an ordinary word. Preserve it even if it resembles an English noun. Give users a translated description next to it rather than inside it. In METS XML structural maps for digital archives, a useful illustration is <structMap TYPE='physical'>. The map represents a physical order in this example. This illustrates why a translated label needs more than a plausible dictionary equivalent: changing physical to logical can alter navigation interpretation. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
2. Hierarchy: a principle that can be tested
Parent-child and ordered relationships determine what a value belongs to. A well-written description must not turn a child component into a top-level one, or describe an attached value as applying globally. In METS XML structural maps for digital archives, a useful illustration is <div ID='chap02' LABEL='Chapter Two' TYPE='chapter'>. LABEL is a human-readable display name. This illustrates why a translated label needs more than a plausible dictionary equivalent: changing ID while translating LABEL breaks references. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
3. Type: a principle that can be tested
A stored value has a type and often a unit or code system. Translation must not coerce integer to text, an identifier to a number, or an encoded role into a familiar but different role. In METS XML structural maps for digital archives, a useful illustration is <div ORDER='10' LABEL='Plate X'>. ORDER expresses a structural order where used. This illustrates why a translated label needs more than a plausible dictionary equivalent: lexicographic sorting puts 10 before 2. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
4. Scope: a principle that can be tested
Distinguish a property of the complete object from a property of one component. Say explicitly what the rule governs when an abbreviation can apply in several positions. In METS XML structural maps for digital archives, a useful illustration is <fptr FILEID='IMG0009'/>. Pointer connects a division to a file entry. This illustrates why a translated label needs more than a plausible dictionary equivalent: translating IMG0009 destroys lookup. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
5. Time and order: a principle that can be tested
Sequence, timestamps, offset and scheduling information are often operational. A natural-language expansion must preserve event order and avoid inventing a time zone, local date or precision absent from the source. In METS XML structural maps for digital archives, a useful illustration is <FLocat xlink:href='images/p009.tif'/>. Location refers to a stored file. This illustrates why a translated label needs more than a plausible dictionary equivalent: editing path to a translated filename causes 404 errors. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
6. Human interface: a principle that can be tested
The translator usually controls captions, tooltips, help text, glossaries and interpretive notes. Those are legitimate spaces for clarity and localization, provided each text is mapped to the correct source key. In METS XML structural maps for digital archives, a useful illustration is <div ADMID='ADM2'>. ADMID joins administrative metadata to structure. This illustrates why a translated label needs more than a plausible dictionary equivalent: treating rights metadata as page text loses its role. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
7. Evidence: a principle that can be tested
A claim should be traceable to a defined field, official standard or verified implementation. If it cannot be proven, say what remains unverified and do not launder a guess into the published wording. In METS XML structural maps for digital archives, a useful illustration is <area BETYPE='TIME' BEGIN='00:06:18' END='00:10:03'/>. A sub-file area denotes a timed segment. This illustrates why a translated label needs more than a plausible dictionary equivalent: converting times into page numbers invents navigation. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
8. Reversibility: a principle that can be tested
A reviewer should be able to trace each translated decision to the original field and reconstruct what was preserved. Without source alignment, corrections become guesswork and later revisions risk data loss. In METS XML structural maps for digital archives, a useful illustration is PROFILE='institution-profile'. Repository profile sets additional constraints. This illustrates why a translated label needs more than a plausible dictionary equivalent: assuming all METS profiles allow identical div/fptr patterns is unsafe. A reviewer should ask the operator to point to the exact source evidence, explain the relationship in ordinary language, and show that the edited human-facing text has not modified the represented fact.
The practical test is to compare the source, the edited copy and the rendered explanation side by side. Record which properties were intentionally localized, which properties were protected and how the result was checked. If a team cannot answer all three, it has not finished quality assurance for this principle. That test generalizes to multilingual training materials, technical help centres and public-facing records.
A repeatable twelve-stage translation workflow
Use this workflow for small one-line notices and larger documentation releases. It is deliberately sequential: verify the assignment and inputs before translating, examine the graph of relationships before approving wording, and leave a small evidence trail that the next editor can reuse. Each step should produce a tangible artifact, not merely an assertion that a check was done.
Step 1: Define the assignment and its exclusions
Before translating a line, state whether the assignment covers viewer labels, human guidance, a source-data object, or actual protocol conversion. A human-readable edition is not authority to rewrite an interoperable file. Ask who will read it, which application will render it, and which elements must remain byte-for-byte unchanged. For METS XML structural maps for digital archives, place <div TYPE='page' LABEL='Page 9'> in the working register and ask whether tYPE classifies the division. The immediate quality checkpoint is: use profile-supplied vocabulary; translate display label instead. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 2: Freeze the source and record its version
Capture an immutable copy and a reference to the relevant specification or application profile. Note which edition, directory, version and implementation produced the data, because an apparently familiar identifier may be given different restrictions by a project-specific profile. Work in a separate annotation or documentation layer. For METS XML structural maps for digital archives, place <div ORDER='10' LABEL='Plate X'> in the working register and ask whether oRDER expresses a structural order where used. The immediate quality checkpoint is: compare numeric sequence without changing assigned order. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 3: Identify the reader’s question
A newcomer asks what happened, what this label means and what action is safe. A maintainer asks which key, object, value or timestamp it refers to. Draft the answer for the actual decision the reader faces, not as a dictionary dump. Prefer a concise direct label followed by an expanded explanation. For METS XML structural maps for digital archives, place <div ORDERLABEL='iv'> in the working register and ask whether order label can show printed roman numbering. The immediate quality checkpoint is: preserve printed pagination and document any user-facing expansion. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 4: Build an invariant register
List control characters, field names, machine identifiers, checksums, typed values, references, ordering and other source facts that must survive unchanged. Put every invariant in a table with its purpose and a verification method. The register prevents well-intentioned translators from silently normalising significant bytes. For METS XML structural maps for digital archives, place <fptr FILEID='IMG0009'/> in the working register and ask whether pointer connects a division to a file entry. The immediate quality checkpoint is: resolve FILEID against fileSec before release. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 5: Draw relationships before choosing words
Trace parent-child relationships, cross-references, message flows, numeric ranges and what each field points at. A translation can be grammatically elegant yet wrong if it attaches a quantity to the wrong object or explains a reference as content. Make a small diagram or two-column map before the first draft. For METS XML structural maps for digital archives, place <fileGrp USE='master'> in the working register and ask whether uSE groups a rendition by local convention. The immediate quality checkpoint is: follow repository profile and translate explanatory description. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 6: Classify every text span
Mark fields as stable machine code, translatable human description, potentially translatable free text, identifier, numeric data, controlled vocabulary or uncertain. Do not infer that a value is translatable merely because it contains English-looking letters. Request authority for the uncertain class. For METS XML structural maps for digital archives, place <FLocat xlink:href='images/p009.tif'/> in the working register and ask whether location refers to a stored file. The immediate quality checkpoint is: keep link exact and test path resolution. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 7: Translate the operational meaning
Write a natural explanation of who does what, to which object, under which condition and at which stage. Keep quantities, polarity, role and scope explicit. If two target-language terms seem possible, choose the one that preserves the process distinction and document why. For METS XML structural maps for digital archives, place <div DMDID='DMD09'> in the working register and ask whether dMDID connects the division to descriptive metadata. The immediate quality checkpoint is: validate references after label localization. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 8: Make an aligned bilingual mapping
Use source key, proposed label, context, justification, owner and status fields. This separates the immutable source value from the display phrase. For each name, link back to a unique identifier so reviewers can discuss meaning without corrupting the original data. For METS XML structural maps for digital archives, place <div ADMID='ADM2'> in the working register and ask whether aDMID joins administrative metadata to structure. The immediate quality checkpoint is: document policy wording separately and preserve ADMID. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 9: Run structural and semantic checks separately
Structural review asks whether identifiers, paths, order and machine types still match. Semantic review asks whether the translated words preserve meaning, roles, distinctions and expected actions. Neither test replaces the other. A file can parse and still mislead; a phrase can read well and still break the workflow. For METS XML structural maps for digital archives, place <mptr xlink:href='issue12.xml'/> in the working register and ask whether pointer can refer to another METS document. The immediate quality checkpoint is: check target record and describe what it connects. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 10: Exercise realistic edge cases
Test missing values, duplicates, nested objects, unusual language directions, empty fields and out-of-order observations. Include a case where the value looks ordinary but is a control token. Record what a renderer, reader or downstream operator actually does rather than assuming it from the translated prose. For METS XML structural maps for digital archives, place <area BETYPE='TIME' BEGIN='00:06:18' END='00:10:03'/> in the working register and ask whether a sub-file area denotes a timed segment. The immediate quality checkpoint is: preserve BETYPE and exact begin/end boundaries. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 11: Review with a second discipline
Ask one language reviewer to focus on clarity and one domain reviewer to focus on technical identity. Present each reviewer with source, translation, invariants and evidence. An expert who sees only the translated paragraph may miss a link that was silently severed. For METS XML structural maps for digital archives, place <seq><area FILEID='F1'/><area FILEID='F2'/></seq> in the working register and ask whether an ordered series binds related resource segments. The immediate quality checkpoint is: check reading order alongside source chronology. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Step 12: Release a traceable edition and transfer learning
Publish the display translation or explanatory guide with version, source reference, unresolved questions and test outcomes. Keep a glossary limited to this domain and a regression suite of representative cases. When a source revision arrives, compare changed fields rather than blindly rewriting every surviving label. For METS XML structural maps for digital archives, place PROFILE='institution-profile' in the working register and ask whether repository profile sets additional constraints. The immediate quality checkpoint is: validate chosen local profile rather than exporting an invented universal rule. Do not advance this field simply because a fluent translation has been proposed.
A reusable deliverable for this step is a compact row containing the original value, its operational role, the target-language label, a link to the specification, and the reviewer decision. Such rows are small enough to audit and rich enough to prevent an unresolved assumption from disappearing during editing. In later projects, this record also supports training and regression tests.
Worked examples: interpret, translate, verify
The following scenarios are constructed teaching examples. They show how to read a source field, identify what may be translated and detect a mistaken interpretation. They are not authenticated shipping transactions, archive records, production image files or live control messages. Always check the precise standard edition and local implementation for a real submission.
Worked example 1: structMap TYPE
Source specimen: <structMap TYPE='physical'>. What it establishes: The map represents a physical order in this example. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Changing physical to logical can alter navigation interpretation. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Preserve controlled attribute values and translate UI legend separately. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 2: logical hierarchy
Source specimen: <structMap TYPE='logical'>. What it establishes: Nested divisions describe intellectual sections. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Flattening chapters into pages changes reader understanding. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Trace nested div tree before translating labels. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 3: div LABEL
Source specimen: <div ID='chap02' LABEL='Chapter Two' TYPE='chapter'>. What it establishes: LABEL is a human-readable display name. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Changing ID while translating LABEL breaks references. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Edit text of LABEL only, preserving stable identifiers. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 4: div TYPE
Source specimen: <div TYPE='page' LABEL='Page 9'>. What it establishes: TYPE classifies the division. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Localising a controlled TYPE token may violate a local profile. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Use profile-supplied vocabulary; translate display label instead. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 5: ORDER field
Source specimen: <div ORDER='10' LABEL='Plate X'>. What it establishes: ORDER expresses a structural order where used. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Lexicographic sorting puts 10 before 2. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Compare numeric sequence without changing assigned order. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 6: ORDERLABEL field
Source specimen: <div ORDERLABEL='iv'>. What it establishes: Order label can show printed roman numbering. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Converting iv into 4 changes source pagination convention. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Preserve printed pagination and document any user-facing expansion. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 7: fptr FILEID
Source specimen: <fptr FILEID='IMG0009'/>. What it establishes: Pointer connects a division to a file entry. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Translating IMG0009 destroys lookup. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Resolve FILEID against fileSec before release. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 8: fileGrp USE
Source specimen: <fileGrp USE='master'>. What it establishes: USE groups a rendition by local convention. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Confusing preservation masters with access copies changes policy. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Follow repository profile and translate explanatory description. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 9: file FLocat
Source specimen: <FLocat xlink:href='images/p009.tif'/>. What it establishes: Location refers to a stored file. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Editing path to a translated filename causes 404 errors. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Keep link exact and test path resolution. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 10: dmdSec association
Source specimen: <div DMDID='DMD09'>. What it establishes: DMDID connects the division to descriptive metadata. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Substituting translated value for DMD09 unlinks descriptive record. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Validate references after label localization. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 11: amdSec association
Source specimen: <div ADMID='ADM2'>. What it establishes: ADMID joins administrative metadata to structure. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Treating rights metadata as page text loses its role. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Document policy wording separately and preserve ADMID. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 12: mptr external record
Source specimen: <mptr xlink:href='issue12.xml'/>. What it establishes: Pointer can refer to another METS document. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Rendering issue12.xml as a display label conceals the link. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Check target record and describe what it connects. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 13: area BETYPE
Source specimen: <area BETYPE='TIME' BEGIN='00:06:18' END='00:10:03'/>. What it establishes: A sub-file area denotes a timed segment. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Converting times into page numbers invents navigation. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Preserve BETYPE and exact begin/end boundaries. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 14: seq child sequence
Source specimen: <seq><area FILEID='F1'/><area FILEID='F2'/></seq>. What it establishes: An ordered series binds related resource segments. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Reordering F2 before F1 misaligns text and audio. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Check reading order alongside source chronology. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Worked example 15: profile-specific rules
Source specimen: PROFILE='institution-profile'. What it establishes: Repository profile sets additional constraints. Start by identifying the object’s scope and the authority for interpreting this field. Describe what a new reader should learn from it before deciding where to put the translated words. The target text should explain the function of the value, not silently convert the value into a different kind of evidence.
Tempting but wrong interpretation: Assuming all METS profiles allow identical div/fptr patterns is unsafe. The error may look harmless in a glossary because the surrounding words remain natural. In an actual workflow, however, the affected party could sort, route, render, approve or reject something based on the wrong assumption. Keep the source specimen as a frozen reference; if an example is hypothetical, make that status visible to the reader rather than implying it comes from a verified transaction.
Correct translation decision: Validate chosen local profile rather than exporting an invented universal rule. The editor should preserve the technical key in the data and provide a separate human-readable explanation naming the field’s role. A domain reviewer then confirms the explanation from the relevant standard and the local producer’s implementation. If the specification alone does not settle the question, the correct output is a qualified note and a request for the missing profile, not an invented rule.
Quality check and transfer: ask a second reader to locate the same record with only the identifiers available; ask a language reviewer to paraphrase the human explanation without seeing the first draft; and ask a technical reviewer to state one realistic failure that the translation must prevent. If the two reviewers disagree, reopen this example and document the specific claim under dispute. The lesson transfers to any multilingual structured record whose labels are easier to change than its underlying relationships.
Common failure modes and their repairs
Failure 1: orderlabel field is misread
A characteristic failure begins with <div ORDERLABEL='iv'>: converting iv into 4 changes source pagination convention. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: preserve printed pagination and document any user-facing expansion. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 2: fptr fileid is misread
A characteristic failure begins with <fptr FILEID='IMG0009'/>: translating IMG0009 destroys lookup. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: resolve FILEID against fileSec before release. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 3: filegrp use is misread
A characteristic failure begins with <fileGrp USE='master'>: confusing preservation masters with access copies changes policy. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: follow repository profile and translate explanatory description. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 4: file flocat is misread
A characteristic failure begins with <FLocat xlink:href='images/p009.tif'/>: editing path to a translated filename causes 404 errors. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: keep link exact and test path resolution. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 5: dmdsec association is misread
A characteristic failure begins with <div DMDID='DMD09'>: substituting translated value for DMD09 unlinks descriptive record. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: validate references after label localization. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 6: amdsec association is misread
A characteristic failure begins with <div ADMID='ADM2'>: treating rights metadata as page text loses its role. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: document policy wording separately and preserve ADMID. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 7: mptr external record is misread
A characteristic failure begins with <mptr xlink:href='issue12.xml'/>: rendering issue12.xml as a display label conceals the link. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: check target record and describe what it connects. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 8: area betype is misread
A characteristic failure begins with <area BETYPE='TIME' BEGIN='00:06:18' END='00:10:03'/>: converting times into page numbers invents navigation. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: preserve BETYPE and exact begin/end boundaries. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 9: seq child sequence is misread
A characteristic failure begins with <seq><area FILEID='F1'/><area FILEID='F2'/></seq>: reordering F2 before F1 misaligns text and audio. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: check reading order alongside source chronology. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Failure 10: profile-specific rules is misread
A characteristic failure begins with PROFILE='institution-profile': assuming all METS profiles allow identical div/fptr patterns is unsafe. A surface review may not catch it because the rewritten English still reads smoothly. Repair the problem by returning to field scope and the source version, then apply the discipline: validate chosen local profile rather than exporting an invented universal rule. Keep a before-and-after snapshot of the label and evidence that the protected data did not change.
The prevention test is concrete: give a reviewer the source object and ask which decision would differ if the flawed wording were trusted. Record that answer as a regression condition. A useful quality checklist tracks severity separately from fluency; an incorrect routing, order, unit, link or execution condition must remain a blocker even when the phrasing seems natural.
Practice workshop: independent decisions and explained answers
Try each problem without looking at its solution. Make a two-column note: what is stable source data, and what may be changed for clarity? Then describe the test you would use to prove that the answer is right. The exercise is not to produce the prettiest sentence; it is to preserve an actionable meaning while showing your work.
Exercise 1: logical hierarchy
Task: Your colleague proposes to translate or normalize <structMap TYPE='logical'> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: Nested divisions describe intellectual sections. Therefore trace nested div tree before translating labels. An unacceptable shortcut would be flattening chapters into pages changes reader understanding. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 2: div TYPE
Task: Your colleague proposes to translate or normalize <div TYPE='page' LABEL='Page 9'> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: TYPE classifies the division. Therefore use profile-supplied vocabulary; translate display label instead. An unacceptable shortcut would be localising a controlled TYPE token may violate a local profile. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 3: ORDERLABEL field
Task: Your colleague proposes to translate or normalize <div ORDERLABEL='iv'> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: Order label can show printed roman numbering. Therefore preserve printed pagination and document any user-facing expansion. An unacceptable shortcut would be converting iv into 4 changes source pagination convention. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 4: fileGrp USE
Task: Your colleague proposes to translate or normalize <fileGrp USE='master'> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: USE groups a rendition by local convention. Therefore follow repository profile and translate explanatory description. An unacceptable shortcut would be confusing preservation masters with access copies changes policy. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 5: dmdSec association
Task: Your colleague proposes to translate or normalize <div DMDID='DMD09'> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: DMDID connects the division to descriptive metadata. Therefore validate references after label localization. An unacceptable shortcut would be substituting translated value for DMD09 unlinks descriptive record. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 6: mptr external record
Task: Your colleague proposes to translate or normalize <mptr xlink:href='issue12.xml'/> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: Pointer can refer to another METS document. Therefore check target record and describe what it connects. An unacceptable shortcut would be rendering issue12.xml as a display label conceals the link. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 7: seq child sequence
Task: Your colleague proposes to translate or normalize <seq><area FILEID='F1'/><area FILEID='F2'/></seq> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: An ordered series binds related resource segments. Therefore check reading order alongside source chronology. An unacceptable shortcut would be reordering F2 before F1 misaligns text and audio. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Exercise 8: structMap TYPE
Task: Your colleague proposes to translate or normalize <structMap TYPE='physical'> in a training guide. Identify one protected element, draft an accurate explanation for an unfamiliar reader, and specify the minimum evidence needed before publication. Imagine that the recipient will act on the guide tomorrow without access to the original editor.
Model reasoning: The map represents a physical order in this example. Therefore preserve controlled attribute values and translate UI legend separately. An unacceptable shortcut would be changing physical to logical can alter navigation interpretation. The evidence must include the source occurrence, a verified field definition, the selected target-language wording and a test showing unchanged relationships. If a required assumption remains unverified, label it as unresolved and block that particular assertion from a definitive release.
Frequently asked questions
Should I translate the codes themselves?
Usually not when they are defined as machine identifiers, syntax tokens or references. First check the schema and local implementation. Translate the associated explanation instead. Even an English-looking key may be used by software as an exact-match string, so a good translator treats the code as evidence to preserve rather than prose to improve. For a concrete check here, consider <structMap TYPE='physical'>: preserve controlled attribute values and translate UI legend separately.
Can a translation be linguistically correct but technically wrong?
Yes. An elegant sentence can associate a value with the wrong party, stage or file. The practical distinction is between grammatical acceptability and operational equivalence. Both require review. For technical records the latter also means stable identifiers, cross-references, units, scope, ordering and conditions, not just reasonable wording. For a concrete check here, consider <div TYPE='page' LABEL='Page 9'>: use profile-supplied vocabulary; translate display label instead.
What if the source is missing a local implementation profile?
Do not pretend the public standard defines every local convention. Preserve raw values, translate clearly identified general terms and mark profile-dependent explanations as provisional. Ask the owning team to supply the profile or a validated example. Do not let an unresolved uncertainty turn into a positive factual assertion through stylistic editing. For a concrete check here, consider <fptr FILEID='IMG0009'/>: resolve FILEID against fileSec before release.
How do I handle terms with no short target-language equivalent?
Use a concise natural description at first mention, followed by the stable source term or code when helpful. In dense tables, prefer a short approved label with a nearby glossary entry. The goal is stable comprehension; inventing a terse but false synonym harms both learners and experts. For a concrete check here, consider <div DMDID='DMD09'>: validate references after label localization.
Can AI translation tools process this format directly?
They can assist with explanatory prose, but a generative output is not a structural validator. Extract permitted text spans, keep machine keys protected, compare source and target fields and conduct domain review. Never infer a safe production change solely from a plausible generated paraphrase. For a concrete check here, consider <area BETYPE='TIME' BEGIN='00:06:18' END='00:10:03'/>: preserve BETYPE and exact begin/end boundaries.
How should a glossary be organised?
Make it key-aware. Record the concept, its source identifier, exact target phrase, prohibited near-synonyms, scope, edition and reviewer. A generic alphabetical glossary without source keys can accidentally merge two different concepts that merely share a similar English word. For a concrete check here, consider <structMap TYPE='physical'>: preserve controlled attribute values and translate UI legend separately.
What counts as final quality assurance?
A credible release contains a source-to-target map, a no-change report for protected fields, a semantic review of material decisions, representative edge-case tests and a named owner for unresolved questions. It should tell a new reviewer where a translation came from and which assumptions were checked. For a concrete check here, consider <div TYPE='page' LABEL='Page 9'>: use profile-supplied vocabulary; translate display label instead.
How should beginners practise without production access?
Use constructed sample records, mark them as educational and compare every field against official public documentation. Practice naming the actor, object, action, time, qualification and evidence limit. Do not fabricate realistic-looking production credentials or assert that an untested example is valid for every implementation. For a concrete check here, consider <fptr FILEID='IMG0009'/>: resolve FILEID against fileSec before release.
Which other translation articles should I read?
Start with the linked series owners for the nearest adjacent skill, then use the Vocabulary hub for precision of meaning and How English Works for sentence-level clarity. The present guide owns only this particular structured-record problem; those routes teach related concepts without repeating the same canonical job. For a concrete check here, consider <div DMDID='DMD09'>: validate references after label localization.
A practical release checklist
- Source and implementation edition recorded.
- Protected keys, structural fields, numeric values and links inventoried.
- Translatable text spans extracted rather than edited blindly in the source object.
- Role, time, scope, unit and hierarchy understood from the correct context.
- Terminology reviewed for false synonyms and unapproved expansions.
- Translated labels mapped back to stable source identifiers.
- Structural checks passed without unintended changes.
- Semantic checks performed by a language reviewer and domain reviewer.
- Ambiguities and local-profile dependencies recorded visibly.
- Final guide published with evidence, version and a transfer plan.
Further reading: authoritative references and next routes
For the technical definitions and version-specific restrictions, use Library of Congress METS overview and tutorial and Library of Congress METS structural map schema documentation. These are the primary references for the present canonical job. Treat vendor examples and local policy as supplementary evidence, not substitutes for a documented implementation profile. Standards and recommendations can evolve; re-check them when a new software or content version is deployed.
To continue learning without confusing adjacent topics, read Dublin Core Metadata and Rights Statements and EPUB Navigation and Reading Landmarks. For general word meaning, morphology and retrieval, visit the eduKateSG Vocabulary Learning Hub. For broader English sentence structure and readable explanation, see How English Works. These routes support the translator’s language judgement while this article retains the separate job of preserving structured technical meaning.
The durable lesson is straightforward. Keep the source object and every protected reference intact, place translated words in the correct human-facing location, and make every important interpretation defensible. The work is complete when a reader can understand the same relevant facts, a downstream system can still interpret the same structure, and a future editor can audit how both outcomes were achieved.