eduKate Singapore

VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Next useful routeCanonical hubHELP

How People Translate Quickly | SBML Level 3 Model Notes: Translate Biology Explanations Without Changing Simulation Equations

—

by

eduKate Asia

To translate SBML model notes quickly, separate the biological explanation meant for a reader from the mathematical model consumed by simulation software. Systems Biology Markup Language (SBML) stores species, compartments, reactions, parameters and equations in a structured XML document. A Spanish or Chinese explanation can help a learner understand that document, but changing a species identifier, kinetic expression, unit definition or initial condition may change what the software computes. The practical method is to extract approved display text, translate it with biological context, and verify that the model’s computational meaning has not moved.

SBML Level 3 translation quality assurance is unusually important because a file can remain valid XML while no longer representing the same biological system. The model may still open, its diagram may still look plausible, and its output may still produce smooth curves. None of those reassuring signs proves that its rates or assumptions survived. A fast, reliable translator therefore keeps an untouched source, makes a localized derivative or an external language guide, and compares machine-readable structure as well as the human explanation.

This guide is for systems-biology students, laboratory educators, translators working with computational biologists and research-data stewards. It answers one specific task: how to localize model titles, descriptive names and approved XHTML notes while preserving an SBML Level 3 model’s identifiers and behavior. It is not a manual for converting mathematical notation or rebuilding a model in another language. The examples show where the safe editing boundary lies, how to diagnose common mistakes, and how to review results without turning translation into a new scientific experiment.

Start with the artifact the audience actually needs

A classroom may need a bilingual explanation of a signaling pathway, not a second scientific model file. If the original SBML is already archived and cited, a translated web page or sidecar handout may be the best product. Students can read the explanation in their preferred language while downloading the untouched source when they need to reproduce published results. This arrangement also avoids the temptation to put multilingual text into fields that were never designed as a language-tagged translation container. Begin by asking who will read the language, in which tool, and at which moment of learning.

If the delivery system genuinely displays the SBML name attribute or the XHTML inside a notes element, a localized copy may be justified. Give it a separate filename, locale identifier, documented relationship to the master and its own checksum. Keep the source URL or accession in the accompanying manifest. The source remains the authority for science; the derivative serves comprehension. This difference is especially important when a model is reused in manuscripts, coursework, repositories and automated simulation pipelines that may disagree about which displayed name is canonical.

The translation brief should say explicitly whether it covers only the model-level description or explanatory notes attached to individual species, compartments and reactions. Those are different scopes with different risks. A species-level note may explain an abbreviation needed to read a pathway diagram, whereas a reaction-level note may mention an experimental assumption or an exception to the model. A blanket request to ‘translate the model’ conceals these distinctions. Define the intended output before opening an editor so reviewers can detect both missing work and unauthorized additions.

Understand the SBML boundary between labels and computation

Most scientific SBML objects have an id that software can use for references and, where allowed, a human-oriented name. A kinetic formula can refer to an identifier while a diagram displays a descriptive name. The model’s id is not a vocabulary exercise: changing it without coordinated reference changes may break a formula or a tool’s connections. In contrast, an approved name often can be made more understandable to a target-language reader. Even then, some downstream software may use names for legend matching, exports or laboratory reconciliation, so labels require an interface check rather than a blanket promise of harmlessness.

The notes element exists for optional material intended for human readers and uses XHTML content under prescribed rules. It is a useful home for an explanation of the model’s assumptions, provenance or biological interpretation. The annotation element serves a different purpose: application-specific structured metadata, sometimes including machine-interpretable biological identifiers and resource links. Do not conclude that an annotation is safe to translate simply because its XML nodes contain words. Where software expects particular resource syntax, modifying that content can disconnect biological meaning from external databases.

There are more protected elements than the equations alone. Identifiers, metaid links, SBO terms, mathematical expressions, units, stoichiometries, reaction reversibility, initial concentrations, event triggers and assignments, function definitions and package-specific constructs may all affect interpretation. A competent handoff marks them as out of scope unless the scientist explicitly commissions a model-editing task. Think of the file as a contract whose explanatory cover can change while its clauses remain fixed. Readers should be able to distinguish a genuine language improvement from a scientific revision by examining the recorded differences.

Inventory the whole model instead of searching for English-looking strings

Make a read-only inventory of the source file with its checksum, SBML level and version, enabled packages, software used to parse it and the model’s stable identifier. Record a list of compartments, species, parameters, reactions, events, rules, function definitions and unit definitions. For each object, note whether a name, a notes tree or an annotation exists. The inventory turns an intimidating XML file into a finite translation assignment and gives a reviewer a precise record against which to check coverage.

Each proposed segment should carry an object path or stable id, its kind, the exact source phrase, the target locale, the proposed target text, the responsible reviewer and a disposition. Suggested dispositions include translate, preserve, explain externally, needs scientific clarification and exclude. Record why a phrase is excluded, especially when the text resembles ordinary English but is actually an identifier or ontology code. This helps a second translator understand the protected boundary and prevents a later round from undoing a careful earlier decision.

Avoid a spreadsheet whose only keys are ‘row 1’ and ‘row 2’. XML element order can change when tools serialize a file, and two reactions may share similar display names. A stable object identifier plus field role is safer than a screen coordinate or a line number. If the model contains anonymous or ambiguously named components, assign temporary inventory IDs in the translation worksheet rather than inserting permanent new IDs into the SBML master. That preserves traceability without making a language review alter the research artifact.

Distinguish names, identifiers and ontology references in practice

Imagine a model with a species whose identifier is S_glucose_ext and whose visible name is ‘Extracellular glucose’. A natural Chinese label might describe glucose outside the cell, but the underlying identifier must remain unchanged in this localization job. If a reaction formula references S_glucose_ext, silently replacing that token with Chinese characters is not a translation; it is an unreviewed model refactor. A viewer can show a translated label in a legend while formulas continue to use the original stable symbol.

A species may also carry an SBO term or a resource annotation linking it to a controlled biological concept. These identifiers are helpful precisely because different languages can refer to the same concept without rewriting its identity. Use them as context while translating: confirm whether the species is a chemical pool, an enzyme complex or an observable before choosing a target-language term. Do not modify the term code to make it ‘look local’. If the annotation itself contains descriptive prose from a third-party source, examine that particular schema and its rights separately rather than applying a generic global replacement.

Sometimes the English display name is a historical abbreviation that a research group intentionally preserves. In that case a bilingual glossary may serve readers better than replacing the canonical short name everywhere. A course can display ‘ERK (extracellular signal-regulated kinase)’ in a first-use explanation and keep ERK in legends. A translator should ask whether consistency across slides, plots and exercises matters more than translating each word. Reproducibility has a social component: people must still recognize that two views describe the same modeled entity.

Treat XHTML notes as documents inside a document

SBML notes are structured XHTML rather than arbitrary bytes after an XML tag. Paragraphs, emphasis and links can carry meaning, and the XML parser must recognize the XHTML namespace. When translating them, keep element structure intact unless a documented renderer requires a justified change. A list may remain a list even when the language changes sentence length. A superscripted reference or mathematical symbol should retain its position. A careful extraction process provides translators with plain readable segments and gives the reassembler enough context to restore safe markup.

A notes paragraph might say, ‘The reaction is assumed to be irreversible under the conditions described in the paper.’ That is a statement about the model’s scientific assumption. A translation that drops ‘assumed’ or adds ‘always’ changes epistemic strength, even if no rate constant is modified. Reviewers should mark uncertainty words, conditions, scope limitations and citations before polishing the sentence. Translation accuracy is not only a matter of correct noun equivalents; it also requires keeping the level of confidence and the boundary of the claim.

If a note includes a source citation, preserve the author, year, DOI and link target independently of the language surrounding it. A translator may translate ‘see Figure 2’ into the target language, but should check that the cited figure number remains correct and that the hyperlink still points to the intended publication. Do not invent a newer reference because it sounds more helpful. Scientific notes often live for years after their original software and publication context. Clarity comes from making that context visible rather than replacing it with contemporary claims that were not part of the model.

Use a minimal worked XML example as a teaching boundary

Consider a compact SBML Level 3 Version 2 example with a model identifier, a descriptive name and human-readable notes. The following illustrates the protected and translatable surfaces; it is not meant to stand in for a full tested kinetic model with species and reactions. The id denotes machine identity, the name serves display purposes, and the XHTML paragraph explains how a reader should understand the model. The valid XML structure must survive whether the target sentence is one line or several.

<?xml version="1.0" encoding="UTF-8"?>
<sbml xmlns="http://www.sbml.org/sbml/level3/version2/core"
      level="3" version="2">
  <model id="glucose_illustration" name="Glucose teaching model">
    <notes>
      <body xmlns="http://www.w3.org/1999/xhtml">
        <p>This model illustrates a simplified glucose uptake example.</p>
      </body>
    </notes>
  </model>
</sbml>

A French learning copy could change the human-facing model name to ‘Modèle pédagogique du glucose’ and the paragraph to ‘Ce modèle illustre un exemple simplifié d’absorption du glucose.’ The identifier, SBML version, namespace and XML element names remain unchanged. The target sentence is intentionally cautious: it still says ‘illustrates’ and ‘simplified’ rather than claiming this minimal teaching artifact contains a validated full model. Production translators would also ask a French-speaking biologist whether the preferred term in this context should be ‘captation’, ‘absorption’ or another domain-specific expression.

The example shows what a small safe change looks like, not what is safe in every file. A laboratory may rely on name values in scripts despite the field being human-oriented in the standard. Therefore compare exported diagrams and tool behavior as well as XML structure. The edit is approved only when its target audience benefits and the researcher accepts its effect on interfaces. A standards-permitted label edit is not the same as an operationally risk-free edit in an undocumented local pipeline.

Protect mathematical expressions and numerical assumptions explicitly

Reaction rate formulas, assignment rules and event expressions are encoded in mathematical structures such as MathML. Their symbols are tied to IDs and model semantics. Translation software must not search-and-replace within those expressions as though every text node is a prose fragment. A rate law that contains a named parameter can be displayed with a translated explanation alongside it; the parameter symbol and numerical value remain protected. Changing a decimal separator, multiplication operator or case of a symbol can make an expression invalid or, worse, valid but different.

The same restriction applies to initial concentrations, compartment sizes and units. A tutorial may explain that a variable represents a concentration in a given unit, but the source’s unit definitions are mathematical data, not language choices. Do not convert seconds to minutes merely because a target-language textbook usually discusses minutes. Unit conversion would require a scientifically reviewed model transformation affecting constants and equations; it cannot be hidden inside a translation task. Keep the original values and provide human-readable units in a separate explanatory layer when helpful.

SBML events, constraints and package features add less obvious hazards. A sentence in an XHTML note may summarize the condition for an event, while the machine condition is implemented elsewhere. If the two disagree after translation, learners will trust the wrong account of the dynamics. A reviewer should trace each consequential claim in the prose to the corresponding model object or source paper. This does not mean altering code to match a nice sentence. It means correcting the sentence so that the learner receives the same system the simulation implements.

Build a glossary around biological function, not single words

A research glossary should identify each term with its modeled role, evidence source, accepted synonym and first-use rule. ‘Activation’ can describe a reaction, a measured increase in activity or a broader biological process. ‘Expression’ may mean gene expression rather than verbal expression. ‘Pool’ can represent a modeled amount or concentration compartment without referring to a physical pool. Translating such terms independently creates plausible sentences that disagree with each other or with the simulation’s assumptions.

Create an entry for each recurring source phrase and link it to the model objects where it occurs. Include the chosen target-language rendering, any abbreviation preserved verbatim and a short note about what it does not mean. For example, distinguish ‘transport into the cell’ from ‘uptake measured by an assay’ when the model uses one as a mechanism and the other as an observation. An educator can then explain the relationship explicitly instead of hoping students infer it from two loosely translated phrases.

The glossary should also distinguish public-facing wording from expert language. A beginner might need ‘how quickly the reaction happens’ before encountering ‘reaction rate’. An advanced learner may require the formal term in the first sentence. Neither choice is universally correct. What matters is that a simplified phrase does not deny the scientific model’s qualifications, and that the formal symbol remains traceable. Quick translation becomes possible when these reader-level decisions are made once and reused consistently rather than improvised for every paragraph.

Use a controlled extraction and reinsertion workflow

Export candidate display text with each object’s location and a narrow whitelist of fields. One project may approve only model-level name and notes; another may also approve species and reaction names. The extractor should not walk every XML text node and call it a sentence. It must identify role, namespace and ancestry before assigning a segment ID. Store the original segment verbatim and record the XML path separately. If a tool exports rich text as raw XML, warn translators that angle brackets and entity references are structural characters, not stylistic decorations.

Translate in a worksheet or localization platform that keeps protected placeholders visible but immutable. Protect model identifiers, URLs, citations, chemical formulas and special mathematical symbols. Where the target language uses a different sentence order, let the translator rearrange genuine prose while leaving placeholder identities intact. Store unresolved scientific queries alongside the segment rather than inserting guessed definitions into the target. A language reviewer and a model owner can resolve a concentrated query list faster than an uncontrolled series of email corrections spread across dozens of files.

Reinsert approved segments into a copy using an XML-aware library. Avoid regular expressions that replace text between arbitrary tags, since nested markup and namespaces make those patterns fragile. Reparse the output after writing it and compare an allowlist-based diff against the original. If any unapproved attribute, mathematical node or numeric value has changed, fail the localization build. A successful build produces both a readable language view and a machine-verifiable report showing exactly which human-facing fields changed.

Preserve provenance and interpret translations as editions

A translator may discover an ambiguous sentence or a likely error in the original. That is valuable feedback, but an ambiguous explanation does not authorize a scientific correction. Flag it for the model author, record the question and retain the established text until the author resolves it. If the author decides a rate equation was wrong, that change belongs in a separately versioned model revision. Combining scientific repair with translation makes it impossible to attribute a changed simulation result to the correct cause.

Every localized derivative should identify its parent model, source revision, target locale, date, translation reviewer and scientific reviewer. Avoid writing claims such as ‘verified equivalent’ unless equivalence tests were actually performed. A minimal educational mockup and a production model have different evidence requirements. For production work, store original and output hashes, validator versions, an allowlist diff and a summary of exported simulation comparisons. A student should be able to identify the trustworthy source even if the translated diagram is copied into an unrelated classroom presentation.

In repositories and learning management systems, the language view may be distributed separately from the model. That is a legitimate design choice, not an incomplete translation. One canonical source model can support several authorized reader guides and glossaries. It reduces the number of scientific files requiring revalidation when a sentence changes and helps teams avoid contradictory model versions in multiple languages. The cost is that educators must preserve the link between the localized explanation and the exact source revision; otherwise readers can quietly drift back into comparing unlike models.

Validate more than XML well-formedness

An XML parser can prove that element boundaries and entities are legal, but it cannot tell whether the SBML model obeys every relevant consistency rule. Run an SBML-aware validator for the selected level, version and packages, and record which checks were enabled. Any new severe validation error after translation is a release blocker. Existing warnings should also be documented rather than silently counted as fresh failures. A model may already contain a limitation unrelated to the language, and responsible QA distinguishes inherited scientific issues from newly introduced damage.

Next, compare the protected portion of the source and localized XML. Record IDs and their references, mathematical expressions, stoichiometries, unit declarations, initial values, event definitions and annotation identity. A byte-for-byte whole-file comparison is too strict because harmless XML serialization choices may reorder attributes. A semantic comparison is more appropriate, provided the team does not dismiss differences outside the allowlist. It should say exactly which fields are permitted to differ; every other deviation deserves investigation rather than an optimistic label such as formatting noise.

Finally, reopen both artifacts in a model-aware editor or simulator supported by the research team. Compare numerical outputs from agreed test cases where simulation is possible, including boundary conditions and a reference time course. Rendering checks should confirm that the intended language appears without hiding units, truncating biological labels or disconnecting labels from nodes. An identical simulation trace alone cannot prove the translation is semantically correct, just as a flawless translated note cannot prove model equivalence. Both kinds of evidence are necessary for an honest release claim.

Diagnose a model that opens but produces changed results

If the localized model parses successfully but its output curves differ, stop treating the problem as linguistic. Compare the source and derivative at the first model object whose protected representation changed. Look for an identifier that a formula no longer resolves, a numeric field altered by locale-aware formatting, a rate expression rewritten by a generic XML editor, or a package attribute dropped during serialization. Reproducing a single expected trajectory is more useful than debating whether two XML files look alike. Save the parameters and environment used for that comparison so another team member can repeat it.

A second failure pattern is a curve that appears numerically similar but has different legend labels. That may be harmless, or it may conceal a swapped mapping between labels and species. Verify which stable identifier each plotted series represents before approving a localized figure. Avoid comparing images solely by appearance; an incorrect legend can turn a good numerical plot into a false biology lesson. The repair is to rebuild the display mapping from stable IDs, not to edit the scientific curve until it matches a preferred narrative. Explanatory text follows evidence, never the reverse.

If the model uses optional SBML packages, an application may not understand every component even though the source is standards-compliant for its declared package set. Document the software’s package support and avoid claiming your translated file is universally interoperable. A language derivative does not fix compatibility between simulation engines. Report which reader and simulator were tested, which features were skipped and whether the original shows the same limitations. This boundary is especially important when students infer that a failed run proves a translated term was biologically wrong.

Diagnose broken formatting or disappeared notes

When a note disappears after export, inspect the XHTML namespace, encoding and the target tool’s rendering behavior before editing words again. Some programs import SBML science faithfully while displaying only a subset of supported XHTML elements. A valid note may therefore be invisible in one interface and readable in another. Test the same derivative in a standards-aware view and record the difference. If an unsupported element caused the display problem, simplify the approved markup in a controlled way while preserving paragraphs, citations and scientific qualifications; do not strip all structure indiscriminately.

If accented characters appear as replacement symbols, examine the UTF-8 path from the translation worksheet to the generated XML. A code point can be mangled by an intermediate legacy text editor or by an incorrectly declared encoding. Check the raw saved output and the parser’s decoded text separately. A visually plausible symbol is not sufficient when a Greek letter, subscript or unit can change interpretation. The repair should happen at the encoding boundary where corruption occurs, followed by re-extraction and a new comparison against the original source segment.

Longer target-language names may overlap species labels in a pathway diagram. Shortening every name aggressively is not the best first answer because important distinctions may be lost. Consider whether the viewer offers tooltips, wrapping, abbreviation rules or a bilingual legend. A short diagram label can coexist with a full accessible explanation linked by identifier. Evaluate the diagram at ordinary screen sizes, in a printed handout and with the font used by the institution. Record truncation as a presentation problem to repair; do not treat it as permission to alter the model’s identifiers.

Diagnose scientifically attractive but misleading prose

One dangerous translation failure is an overconfident causal claim. A source note may say that a parameter was fitted to observations, not directly measured. If the target text calls it an experimentally measured constant, readers may overestimate the evidence. Another failure is erasing a negative condition: ‘the reaction is not assumed to be reversible’ becomes a sentence implying reversibility. Such errors do not show up in an XML validator or a simulation comparison. They require a bilingual reviewer who understands the modeled mechanism and checks each statement’s claim, evidence and qualifiers.

Terminology drift becomes visible when the same component is described as a catalyst in one note and as the consumed substrate in another. Build a relationship check: for each translated reaction explanation, list reactants, products, modifiers and the role claimed by the prose. Compare these against the modeled reaction and any cited publication. The check is not intended to prove every biological interpretation true; it catches direct contradictions introduced by the language layer. Where the source itself is ambiguous, annotate the editorial question and avoid making the target more definite than the author.

A third failure comes from translating local abbreviations according to common dictionary meanings. In computational biology, ‘ER’, ‘R’, ‘Km’ and similar compact labels may have study-specific meanings. The glossary should preserve the original token and record its documented expansion. When a term appears in a figure, model name and course worksheet, use a consistent first-use explanation so students can connect all three. If no authoritative expansion exists, mark it as unresolved rather than manufacturing one. A careful unknown is more educational than a polished but unsupported guess.

A changed-context transfer exercise: from metabolism to signaling

Suppose a team previously translated notes for a simple glucose uptake teaching model. A new collaborator provides a signaling model in which a species ERK_active is produced by a reaction, and an event changes an intervention parameter when a threshold is reached. The new course audience reads Malay rather than French. The brief says: ‘Translate everything students can see, including the event description and molecule labels.’ Decide whether to translate the species id, model name, XHTML notes, event trigger expression, event-facing explanation and controlled ontology references.

The defensible answer separates presentation from execution again, but the inventory is more demanding. Translate approved display names and explanatory notes with specialist terminology review; keep ERK_active, event trigger mathematics, identifiers, ontology terms and parameter values fixed. If the simulator’s legend uses a human name, test its mapping to the original species ID. If an event description says ‘when the observed response passes the threshold’, preserve whether the underlying trigger really refers to a measured response or to a modeled state. The new context requires checking event semantics, not merely reusing the glucose glossary.

The transfer lesson is that the method remains stable while its protected-object inventory changes. In metabolism, compartment units and stoichiometry may be the main hazard; in signaling, timed events, discrete states and model-specific abbreviations may dominate. A translator who only remembers ‘do not change numbers’ will miss symbolic or relational failures. A translator who remembers ‘identify reader-facing fields, preserve executable state, and verify the claim’ can adapt to either model. Write the answer in a change log so a scientific reviewer can approve its scope independently of the target-language fluency.

A release checklist that a second person can actually follow

Before release, ask a reviewer to open the original manifest, count the segments expected for the chosen scope, and account for every accepted, excluded and unresolved segment. They should verify target-language clarity, preserve uncertainty and check that copied citations resolve. Then they should examine the semantic diff: only permitted name or XHTML notes fields may vary. If a protected attribute is missing, if package declarations differ or if the SBML-aware validation detects a new error, the file is not ready. This simple sequence prevents an attractive localized diagram from bypassing technical checks.

The reviewer should next test the actual reading journey: find a species from the translated explanation, locate it in a diagram, identify its stable model object and reproduce one example analysis from the source course. This is a stronger educational check than proofreading a list of sentences in isolation. It reveals missing context, ambiguous abbreviations and labels that the software silently falls back to English. Mark each test as passed, failed, not applicable or untested. ‘Not tested’ must remain visible instead of being quietly converted into ‘passed’.

Finally, confirm provenance and distribution. Are the source version, derivative language and responsible reviewers named? Can readers retrieve the untouched scientific model? Have permissions for translation and redistribution been checked? Does the repository differentiate explanation updates from scientific model changes? If a later model revision arrives, there should be a route to invalidate or review affected translations. The goal is not paperwork for its own sake. The goal is to let another teacher reuse the lesson confidently without guessing whether the language copy is still tied to the research it describes.

Limits: what this workflow cannot prove

Even a perfect allowlist diff cannot prove that the original model reflects nature correctly. A simulation can reproduce its old output and still omit important pathways or rely on uncertain parameters. Translation QA protects the meaning and computational behavior of a particular artifact; it does not validate the underlying scientific hypothesis. For a new scientific claim, educators should look to primary literature, model provenance and domain review. Avoid phrasing a successful technical check as independent biological confirmation.

There are also application-specific limits. A program may export only some model metadata, a visualization may ignore notes, and a third-party annotation namespace may impose its own content rules. An external glossary or accessible lesson may be more portable than attempting to force multilingual labels into every software file. When a technical ecosystem lacks a native language-map feature, a single authoritative model plus localized explanatory companions is often the more maintainable design. State the trade-off so users know why the delivery contains two linked artifacts rather than one magically multilingual XML file.

Summary: protect the science, translate the understanding

SBML localization succeeds when every change has a documented reason and a measurable boundary. First identify whether learners need notes inside the model or a companion resource. Then extract only approved human-facing text, translate biological roles and qualifications accurately, reinsert with an XML-aware process, and compare both the protected model structure and available simulation behavior. Finish with real reader testing, clear provenance and a report of unresolved limitations. Speed comes from a precise scope and reusable checks, not from treating the entire XML document as an ordinary paragraph.

The most valuable result is not simply a French, Malay or Chinese copy that opens successfully. It is a learning experience in which a student can connect a clear sentence to the correct species, equation and experimental assumption without losing the evidence behind it. That combination of language access, technical integrity and scientific humility makes a translation useful long after the first classroom session.

Frequently asked questions

1. Can I translate an SBML species id into another language?

No, not as part of an ordinary localization task. IDs are machine references and may appear in mathematical expressions and tool interfaces. Keep them stable and translate approved human-facing labels or notes instead. A true identifier refactor requires its own coordinated scientific and software review.

2. Is the SBML name attribute always safe to edit?

The standard distinguishes a human-oriented name from an identifier, but local tools may depend on exact name strings for display mapping or exports. Treat names as candidates, not automatic permissions, and compare downstream tool behavior after a change.

Yes. Notes have specific XHTML requirements. Preserve valid namespaces and markup while translating the prose, and verify that the chosen reader renders the result. A plain XML parser cannot establish whether the presentation is useful.

4. Should we translate RDF resource identifiers in annotation?

No. Resource identifiers carry machine-readable identity. Some annotations may contain project-specific descriptive text, but their schemas require separate review. Do not apply a blanket language replacement to structured annotation content.

5. What if a translated scientific term has several plausible equivalents?

Record the alternatives and ask a domain reviewer which one matches the modeled role. A glossary with context and a source citation reduces repeated disagreements. Avoid replacing a precise term with a vague convenient synonym.

6. Does successful XML parsing mean the localized model is equivalent?

No. XML well-formedness says almost nothing about the preservation of kinetics, references, units or simulated outcomes. Run SBML-aware validation, an allowlist diff and relevant simulation comparisons when the model can be executed.

7. Why retain the original SBML if the derivative is valid?

The original anchors scientific provenance, citation and reproducibility. A derivative exists for language access and may rely on a different rendering or distribution environment. Keeping both prevents a translated explanation from silently replacing a research record.

8. Can a teacher translate only model-level notes and leave species labels unchanged?

Yes, if that scope meets the learning task and is recorded transparently. A bilingual guide can explain stable symbols without rewriting each object. The result should be tested with actual students to ensure important terms remain discoverable.

9. What if a translated note contradicts the underlying reaction?

Stop release and compare the source note, the model and its cited evidence. Correct the translation if it introduced the contradiction; refer a pre-existing ambiguity or scientific error to the model owner as a separate issue.

10. Is a model with identical simulation plots fully translated?

Not necessarily. Plots do not check whether notes are omitted, linguistic uncertainty is preserved or captions identify the correct variables. Language review and computational checks answer different questions; both matter.

Authoritative sources