Scientific Figure Caption Translation Guide

Quick answer. To translate scientific figure captions and graph legends correctly, preserve the source’s exact categories, qualifications, relationships and decision consequences before improving the language. This practical guide explains scientific figure caption translation, graph legend translation, error bars, confidence intervals, sample sizes and research terminology with a repeatable method, worked examples, failure analysis, practice questions and routes for further study. It is written for researchers, translators, graduate students, editors and scientific communicators across countries and language pairs.

The difficult part is not finding elegant words. A graph can stay identical while its translated legend changes what the viewer thinks it proves. Standard deviation is not a confidence interval, an association is not proof of causation, and a percentage-point difference is not necessarily a percent change. A good translation remains readable, but it must also make the same distinctions the source makes; otherwise the reader may take an action, record an answer or make an inference that the source did not support. The first three paragraphs therefore address the search task directly: how to translate correctly, how to check equivalence and how to avoid silent changes in meaning.

A figure caption connects marks on a page to a precisely limited scientific claim. The translation must preserve the measured variable, experimental grouping, uncertainty definition, sample size and the boundary of the conclusion. The result should be usable by readers outside the original country without pretending that every jurisdiction, discipline or institution operates identically. Examples below are fictional teaching cases and English-language meaning checks, not official target-language wording. For publication in a particular language, a competent translator and a field reviewer must check the actual translation in context.

The visible problem: fluent language can still carry a false conclusion

A graph can stay identical while its translated legend changes what the viewer thinks it proves. Standard deviation is not a confidence interval, an association is not proof of causation, and a percentage-point difference is not necessarily a percent change. Imagine two competent readers, each using a different language, faced with the same source record. Their sentences may sound natural in both languages. Yet if the first reader treats a category as optional and the second treats it as mandatory, the translations are not equivalent. The error is not style. It is a change to the underlying information or decision.

The practical benchmark is whether a scientist using only the translated figure can recover the same variables, groups, uncertainty interpretation and stated conclusion. That benchmark deliberately separates writing quality from evidence quality. A polished draft can win praise from ordinary proofreaders while concealing a changed denominator, a missing qualifier or the wrong referent. Translation quality must therefore be tested against the meaning of the source and the tasks of the reader, not only against grammatical beauty.

This page owns captions, plotted symbols and uncertainty statements. It does not own entire research papers, statistical model cards or the mechanical conversion of units. Readers looking for adjacent topics should follow the dedicated articles linked later. Keeping each page to one canonical job avoids treating every translation problem as interchangeable and makes it easier to choose the right verification checklist.

First principles: what must survive translation

Functional equivalence

The translated text must perform the same communicative function. A reader of either language should reach the same interpretation of every panel, colour, axis, measurement, comparator and limitation, with no stronger causal or statistical claim introduced. Start with a purpose statement and then check every sentence against that purpose. Natural language may differ significantly between languages while function remains stable.

In scientific figure captions and graph legends, this principle becomes concrete when examining panel identifiers and reading order. Treat this element as an observable test: show a reviewer the source and target, ask them to identify the same distinction, and resolve discrepancies before approving the next stage. The purpose is to transform an abstract promise of accuracy into an explicit check that someone else could reproduce.

Evidence before elegance

Keep what is demonstrated separate from what is inferred. The source is not improved when a translator turns uncertainty into a confident explanation. Mark the evidence anchors first; make prose fluent afterward. When the source is unclear, document the uncertainty and request expert clarification.

In scientific figure captions and graph legends, this principle becomes concrete when examining axes and scales. Treat this element as an observable test: show a reviewer the source and target, ask them to identify the same distinction, and resolve discrepancies before approving the next stage. The purpose is to transform an abstract promise of accuracy into an explicit check that someone else could reproduce.

Controlled differences

Different languages may require changes in word order, grammar, article usage or sentence division. These differences are legitimate only when they do not alter protected relations. A faithful draft may be less literal than a poor translation because it preserves the action or claim rather than its surface shape.

In scientific figure captions and graph legends, this principle becomes concrete when examining variables and units. Treat this element as an observable test: show a reviewer the source and target, ask them to identify the same distinction, and resolve discrepancies before approving the next stage. The purpose is to transform an abstract promise of accuracy into an explicit check that someone else could reproduce.

Scope and attribution

Ask who is speaking, to whom, about what, at what point in time, and under which stated conditions. A translation that preserves nouns but changes scope, actor or time is still incorrect. Check every reference to a person, institution, item, document and prior statement.

In scientific figure captions and graph legends, this principle becomes concrete when examining legend marks and line styles. Treat this element as an observable test: show a reviewer the source and target, ask them to identify the same distinction, and resolve discrepancies before approving the next stage. The purpose is to transform an abstract promise of accuracy into an explicit check that someone else could reproduce.

Reader testing

A bilingual reviewer detects language problems; a subject specialist detects domain mistakes; a real reader detects unexpected interpretation. Use these as complementary tests. If a reader cannot explain the protected condition back in their own words, the translation is not publication-ready.

In scientific figure captions and graph legends, this principle becomes concrete when examining sample size and independence. Treat this element as an observable test: show a reviewer the source and target, ask them to identify the same distinction, and resolve discrepancies before approving the next stage. The purpose is to transform an abstract promise of accuracy into an explicit check that someone else could reproduce.

Version control

Reference materials evolve. A term or rule valid in one edition may be wrong in another. Keep dated source documents and source identifiers. Do not insert future standard wording into an older document merely because it sounds more up to date.

In scientific figure captions and graph legends, this principle becomes concrete when examining error-bar definitions. Treat this element as an observable test: show a reviewer the source and target, ask them to identify the same distinction, and resolve discrepancies before approving the next stage. The purpose is to transform an abstract promise of accuracy into an explicit check that someone else could reproduce.

A repeatable twelve-step method

1. Define the question and intended reader

Begin with the question the reader is actually trying to answer, not the nearest dictionary equivalent. Record audience, use context, risk of misreading and the exact decision the document supports. A translator should be able to state in one sentence what the reader will do with the translated information. Without that sentence, a polished draft can still miss its function.

For this document type, pay particular attention to panel identifiers and reading order. Consider the fictional item “Bars show means ± one standard deviation.”. A correct meaning check would be “Preserve dispersion measured as standard deviation.”. The tempting error is “Substituting standard error exaggerates the impression of precision.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

A robust acceptance test is to remove the original-language text and ask a knowledgeable target-language reader to describe what the translated item permits, states or measures. Then compare that back-explanation with the original. When they differ, restore the source boundary first; do not merely replace an awkward synonym. This small, repeated exercise turns translation from an intuition-driven activity into an auditable learning method.

2. Lock the original source and version

Preserve the original version, publication date, number of items and any referenced attachments before making changes. Identify whether a newer revision supersedes it, because a translation of yesterday’s text must not pretend to be a translation of today’s different document. Record decisions in a simple review log tied to the source version.

For this document type, pay particular attention to axes and scales. Consider the fictional item “Whiskers indicate 95% confidence intervals.”. A correct meaning check would be “Retain interval definition and 95% level.”. The tempting error is “Calling them a 95% range of individual observations is incorrect.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

3. Separate facts, labels, and explanatory prose

Not all words perform the same job. Some words label a category, others record a measurement, others express uncertainty, and still others explain an action. Mark the protected elements before drafting, then choose natural target-language sentences around those anchors. A literal translation of every word is no substitute for retaining their relationships.

For this document type, pay particular attention to variables and units. Consider the fictional item “Error bars represent SEM across six independent experiments.”. A correct meaning check would be “Retain SEM and six independent experiments.”. The tempting error is “Changing independent experiments to six cells alters the denominator.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

4. Build a field-by-field equivalence map

Make a small alignment table that pairs every required source field with the corresponding target field. Do not rely on visual proximity in a new layout: a label can move across a page or screen while its identifier must still point to the same thing. Missing or repeated fields should be flagged before the text goes for fluent editing.

For this document type, pay particular attention to legend marks and line styles. Consider the fictional item “n = 12 participants; 36 observations.”. A correct meaning check would be “Retain 12 participants and 36 observations as different counts.”. The tempting error is “Treating 36 observations as 36 independent participants overstates evidence.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

5. Protect quantifiers, negation and conditionals

Read every not, only, may, must, unless, before, after, fewer, more and at least. These are small expressions with disproportionate influence over the reader’s inference. Ask whether the translated statement rules in and rules out exactly the same cases. If the conditions differ, the document’s operational meaning has changed.

For this document type, pay particular attention to sample size and independence. Consider the fictional item “Panel a: baseline; panel b: after treatment.”. A correct meaning check would be “Preserve a and b mapping and temporal order.”. The tempting error is “Swapping panel labels changes interpretation.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

6. Preserve dates, periods and reference points

A figure without its period or reference point is frequently a different figure. Check date ranges, event anchors, reporting periods and any historical or current version that determines interpretation. Use the target audience’s date format only when the underlying date stays unequivocal; avoid ambiguous two-number dates in cross-border communication.

For this document type, pay particular attention to error-bar definitions. Consider the fictional item “Blue squares denote controls; orange circles denote treated samples.”. A correct meaning check would be “Preserve both hue and shape associations.”. The tempting error is “Relying on colours alone harms monochrome accessibility.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

7. Check numerical content as structured information

Numbers, sample sizes, quantities, signs, percentages, identifiers and coded labels should receive a separate verification pass. They may look easy because they are short, but errors survive ordinary spell-checking. Compare source and target side by side, then ask whether an independent reader would reconstruct the same calculation or categorization.

For this document type, pay particular attention to confidence intervals. Consider the fictional item “Y-axis uses log10 concentration.”. A correct meaning check would be “Retain base-ten logarithmic scaling.”. The tempting error is “Describing equal tick spacing as equal absolute increments misleads.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

8. Keep degree of certainty unchanged

Distinguish observation from interpretation, measured evidence from allegation, and certainty from possibility. Preserve hedges where evidence is limited. An elegant but overconfident target sentence is not an improvement: it changes the degree of confidence that the source asks the reader to place in a claim.

For this document type, pay particular attention to p-values and statistical tests. Consider the fictional item “Expression increased 1.5-fold relative to baseline.”. A correct meaning check would be “Retain the multiplier and reference baseline.”. The tempting error is “Translating as an increase of 150% changes the quantity.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

9. Resolve ambiguity explicitly rather than guessing

If the source itself could mean two different things, translation cannot reliably settle the issue through stylistic preference. Raise a documented question for the author, project owner or qualified subject specialist. If no resolution is available, maintain the ambiguity or label the uncertainty instead of inventing a neat answer.

For this document type, pay particular attention to limits of detection. Consider the fictional item “Response rate rose from 40% to 50%, a ten-percentage-point difference.”. A correct meaning check would be “Keep the absolute percentage-point distinction.”. The tempting error is “Calling it a ten-percent relative increase is incorrect.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

10. Test the translation in its actual display context

Read the draft where people will encounter it: chart, form, PDF, website, mobile screen or museum panel. Inspect line wrapping, order, typography, headings, symbols and accessibility. A text can be grammatically sound but function incorrectly when a warning is detached from the item it qualifies.

For this document type, pay particular attention to logarithmic scales. Consider the fictional item “No statistically significant difference was detected at the stated threshold.”. A correct meaning check would be “Retain lack of detection, not proof of identical groups.”. The tempting error is “No difference exists is stronger than the finding.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

11. Review with a domain specialist and representative reader

Use separate checks for subject accuracy and reader comprehension. A qualified reviewer can confirm specialist terms; a representative target-language reader can reveal unexpected everyday interpretations. Both are useful because technical correctness without intelligibility is insufficient, and fluency without technical correctness is dangerous.

For this document type, pay particular attention to association versus causation. Consider the fictional item “Spearman rho = 0.62; p = 0.03.”. A correct meaning check would be “Retain association statistic and reported p-value.”. The tempting error is “Correlated variables are not automatically causal.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

12. Release with a documented evidence trail

Record source version, translator, reviewer, unresolved questions, final sign-off and correction route. Publish the exact approved version rather than a later unreviewed rewrite. Where updates occur, compare changed portions against the source and review their dependencies so that one repaired sentence does not silently invalidate another.

For this document type, pay particular attention to reference to supplementary data. Consider the fictional item “Scale bar = 20 μm.”. A correct meaning check would be “Keep micro-unit and graphical reference.”. The tempting error is “Changing micrometres to millimetres multiplies scale by a thousand.”. Before going further, identify which individual word, relationship or formatting choice produces that change; write it in the review log so another reviewer can reproduce the same correction.

Worked examples: twenty translation decisions to inspect

All examples in this section are invented for practice. The quoted source phrase is an illustrative English source; the “preserve” field is a faithful English-language meaning benchmark against which translations into any other language can be checked. It is not a certified translation or a rule that every language must use the same words. Use the examples to see where precision lives.

Worked case 1: Standard deviation

Illustrative source: “Bars show means ± one standard deviation.” Meaning to preserve: Preserve dispersion measured as standard deviation.

This item looks manageable because it is short, yet Substituting standard error exaggerates the impression of precision. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Preserve dispersion measured as standard deviation.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 2: Confidence interval

Illustrative source: “Whiskers indicate 95% confidence intervals.” Meaning to preserve: Retain interval definition and 95% level.

This item looks manageable because it is short, yet Calling them a 95% range of individual observations is incorrect. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain interval definition and 95% level.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 3: Standard error

Illustrative source: “Error bars represent SEM across six independent experiments.” Meaning to preserve: Retain SEM and six independent experiments.

This item looks manageable because it is short, yet Changing independent experiments to six cells alters the denominator. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain SEM and six independent experiments.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 4: Sample size

Illustrative source: “n = 12 participants; 36 observations.” Meaning to preserve: Retain 12 participants and 36 observations as different counts.

This item looks manageable because it is short, yet Treating 36 observations as 36 independent participants overstates evidence. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain 12 participants and 36 observations as different counts.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 5: Panel order

Illustrative source: “Panel a: baseline; panel b: after treatment.” Meaning to preserve: Preserve a and b mapping and temporal order.

This item looks manageable because it is short, yet Swapping panel labels changes interpretation. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Preserve a and b mapping and temporal order.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 6: Colour key

Illustrative source: “Blue squares denote controls; orange circles denote treated samples.” Meaning to preserve: Preserve both hue and shape associations.

This item looks manageable because it is short, yet Relying on colours alone harms monochrome accessibility. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

Worked case 7: Log scale

Illustrative source: “Y-axis uses log10 concentration.” Meaning to preserve: Retain base-ten logarithmic scaling.

This item looks manageable because it is short, yet Describing equal tick spacing as equal absolute increments misleads. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

Worked case 8: Fold change

Illustrative source: “Expression increased 1.5-fold relative to baseline.” Meaning to preserve: Retain the multiplier and reference baseline.

This item looks manageable because it is short, yet Translating as an increase of 150% changes the quantity. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain the multiplier and reference baseline.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 9: Percentage points

Illustrative source: “Response rate rose from 40% to 50%, a ten-percentage-point difference.” Meaning to preserve: Keep the absolute percentage-point distinction.

This item looks manageable because it is short, yet Calling it a ten-percent relative increase is incorrect. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Keep the absolute percentage-point distinction.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 10: Non-significant

Illustrative source: “No statistically significant difference was detected at the stated threshold.” Meaning to preserve: Retain lack of detection, not proof of identical groups.

This item looks manageable because it is short, yet No difference exists is stronger than the finding. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain lack of detection, not proof of identical groups.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 11: Correlation

Illustrative source: “Spearman rho = 0.62; p = 0.03.” Meaning to preserve: Retain association statistic and reported p-value.

This item looks manageable because it is short, yet Correlated variables are not automatically causal. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain association statistic and reported p-value.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 12: Scale bar

Illustrative source: “Scale bar = 20 μm.” Meaning to preserve: Keep micro-unit and graphical reference.

This item looks manageable because it is short, yet Changing micrometres to millimetres multiplies scale by a thousand. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Keep micro-unit and graphical reference.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 13: Detection limit

Illustrative source: “Values below the limit of detection are shown as open triangles.” Meaning to preserve: Retain censored status rather than zeros.

This item looks manageable because it is short, yet Replacing undetected with absent overstates certainty. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Retain censored status rather than zeros.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 14: Box plot

Illustrative source: “Box is interquartile range; line is median; whiskers show 1.5 × IQR.” Meaning to preserve: Preserve each graphical definition.

This item looks manageable because it is short, yet Assuming whiskers show minimum and maximum changes meaning. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

Worked case 15: Multiple comparisons

Illustrative source: “Adjusted p-values are shown after correction for multiple testing.” Meaning to preserve: Keep adjusted status and correction statement.

This item looks manageable because it is short, yet Unadjusted p changes the inference. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Keep adjusted status and correction statement.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 16: Time axis

Illustrative source: “Days since randomization, not days since first dose.” Meaning to preserve: Preserve the event anchoring time zero.

This item looks manageable because it is short, yet Changing anchor shifts chronology. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

Worked case 17: Normalization

Illustrative source: “Counts normalized per 1,000 person-years.” Meaning to preserve: Preserve person-time denominator.

This item looks manageable because it is short, yet Changing to per 1,000 persons erases time exposure. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

Worked case 18: Exploratory panel

Illustrative source: “Exploratory subgroup results; interpret with caution.” Meaning to preserve: Keep exploratory qualifier and scope.

This item looks manageable because it is short, yet Writing confirms treatment benefit makes a claim not supported. The reader’s inference depends on the exact distinction stated in the source. A translator should therefore identify the protected information, copy it into a comparison note and only then formulate a target-language version that fits the intended audience. A smooth sentence that erases the original distinction is not a successful translation.

To verify this case, ask a reviewer who has not seen the source to explain the translated text and give a plausible example of when it applies. Compare that explanation with “Keep exploratory qualifier and scope.”. If the reviewer instead infers the distortion identified above, revise the wording or restore a missing label, qualification or reference. Keep the reason for revision: it will help when another page uses the same concept.

Worked case 19: Standard deviation in an abbreviated format

Illustrative source: “Bars show means ± one standard deviation.” Meaning to preserve: Preserve dispersion measured as standard deviation.

Worked case 20: Confidence interval in an abbreviated format

Illustrative source: “Whiskers indicate 95% confidence intervals.” Meaning to preserve: Retain interval definition and 95% level.

Common failure modes and why the usual quick fixes do not work

1. Replacing a category with a familiar but different one

The replacement sounds natural, so surface proofreading may approve it; however the classification boundary has moved. In scientific figure captions and graph legends, a useful warning sign is the scenario “Y-axis uses log10 concentration.”. The protected reading remains “Retain base-ten logarithmic scaling.”; the wrong shortcut is “Describing equal tick spacing as equal absolute increments misleads.”. Check not only the edited sentence but the downstream reader decision it controls.

Repair method: restore the lost information from the authoritative source, record the reason in a reviewer note, check any repeated occurrences, and ask a second reader to describe what the corrected sentence now means. That final check matters because changing a word may not repair a misplaced label, missing section or obsolete version reference.

2. Deleting a negative or condition

A tiny omitted word can invert eligibility, assurance, certainty or user action even when the surrounding sentence remains perfectly grammatical. In scientific figure captions and graph legends, a useful warning sign is the scenario “Expression increased 1.5-fold relative to baseline.”. The protected reading remains “Retain the multiplier and reference baseline.”; the wrong shortcut is “Translating as an increase of 150% changes the quantity.”. Check not only the edited sentence but the downstream reader decision it controls.

3. Letting a template reorder fields

The translation may still include every label but attach one to the wrong row, panel, document or response code. In scientific figure captions and graph legends, a useful warning sign is the scenario “Response rate rose from 40% to 50%, a ten-percentage-point difference.”. The protected reading remains “Keep the absolute percentage-point distinction.”; the wrong shortcut is “Calling it a ten-percent relative increase is incorrect.”. Check not only the edited sentence but the downstream reader decision it controls.

4. Treating uncertainty as weak writing

Editors sometimes remove hedges to sound confident, inadvertently manufacturing evidence not present in the source. In scientific figure captions and graph legends, a useful warning sign is the scenario “No statistically significant difference was detected at the stated threshold.”. The protected reading remains “Retain lack of detection, not proof of identical groups.”; the wrong shortcut is “No difference exists is stronger than the finding.”. Check not only the edited sentence but the downstream reader decision it controls.

5. Changing a unit or reference period silently

Readers see plausible numbers but infer a different quantity or comparison from the original. In scientific figure captions and graph legends, a useful warning sign is the scenario “Spearman rho = 0.62; p = 0.03.”. The protected reading remains “Retain association statistic and reported p-value.”; the wrong shortcut is “Correlated variables are not automatically causal.”. Check not only the edited sentence but the downstream reader decision it controls.

6. Confusing a name with a description

Official identifiers and credited roles should remain identifiable even when descriptive text is translated. In scientific figure captions and graph legends, a useful warning sign is the scenario “Scale bar = 20 μm.”. The protected reading remains “Keep micro-unit and graphical reference.”; the wrong shortcut is “Changing micrometres to millimetres multiplies scale by a thousand.”. Check not only the edited sentence but the downstream reader decision it controls.

7. Assuming a local rule applies everywhere

A target country may use another process, but translation is not permission to rewrite the source as though it operated under that local rule. In scientific figure captions and graph legends, a useful warning sign is the scenario “Values below the limit of detection are shown as open triangles.”. The protected reading remains “Retain censored status rather than zeros.”; the wrong shortcut is “Replacing undetected with absent overstates certainty.”. Check not only the edited sentence but the downstream reader decision it controls.

8. Editing without the source open

Without immediate side-by-side comparison, a reviewer often checks fluency but misses what disappeared. In scientific figure captions and graph legends, a useful warning sign is the scenario “Box is interquartile range; line is median; whiskers show 1.5 × IQR.”. The protected reading remains “Preserve each graphical definition.”; the wrong shortcut is “Assuming whiskers show minimum and maximum changes meaning.”. Check not only the edited sentence but the downstream reader decision it controls.

9. Passing automated QA without reading the meaning

Spell-checkers and token checks can confirm surface features but cannot reliably judge all domain-specific inferences. In scientific figure captions and graph legends, a useful warning sign is the scenario “Adjusted p-values are shown after correction for multiple testing.”. The protected reading remains “Keep adjusted status and correction statement.”; the wrong shortcut is “Unadjusted p changes the inference.”. Check not only the edited sentence but the downstream reader decision it controls.

10. Publishing a revised source through an old translation

An earlier target version may no longer match an updated field, figure, disclaimer or instruction. In scientific figure captions and graph legends, a useful warning sign is the scenario “Days since randomization, not days since first dose.”. The protected reading remains “Preserve the event anchoring time zero.”; the wrong shortcut is “Changing anchor shifts chronology.”. Check not only the edited sentence but the downstream reader decision it controls.

Practical exercises: move from recognition to independent checking

Exercise 1: Percentage points

Start with “Response rate rose from 40% to 50%, a ten-percentage-point difference.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that keep the absolute percentage-point distinction.. A typical wrong answer would calling it a ten-percent relative increase is incorrect..

Transfer task: change the audience from a specialist to a general reader without changing the source fact. You may add a short explanation in a clearly labelled note if the publishing brief allows it, but do not weave an unsupported inference into the translated source. Finish by checking the field or relationship a reader must identify. An exercise is complete only when you can explain why your revised words preserve the same information.

Exercise 2: Non-significant

Start with “No statistically significant difference was detected at the stated threshold.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that retain lack of detection, not proof of identical groups.. A typical wrong answer would no difference exists is stronger than the finding..

Exercise 3: Correlation

Start with “Spearman rho = 0.62; p = 0.03.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that retain association statistic and reported p-value.. A typical wrong answer would correlated variables are not automatically causal..

Exercise 4: Scale bar

Start with “Scale bar = 20 μm.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that keep micro-unit and graphical reference.. A typical wrong answer would changing micrometres to millimetres multiplies scale by a thousand..

Exercise 5: Detection limit

Start with “Values below the limit of detection are shown as open triangles.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that retain censored status rather than zeros.. A typical wrong answer would replacing undetected with absent overstates certainty..

Exercise 6: Box plot

Start with “Box is interquartile range; line is median; whiskers show 1.5 × IQR.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that preserve each graphical definition.. A typical wrong answer would assuming whiskers show minimum and maximum changes meaning..

Exercise 7: Multiple comparisons

Start with “Adjusted p-values are shown after correction for multiple testing.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that keep adjusted status and correction statement.. A typical wrong answer would unadjusted p changes the inference..

Exercise 8: Time axis

Start with “Days since randomization, not days since first dose.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that preserve the event anchoring time zero.. A typical wrong answer would changing anchor shifts chronology..

Exercise 9: Normalization

Start with “Counts normalized per 1,000 person-years.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that preserve person-time denominator.. A typical wrong answer would changing to per 1,000 persons erases time exposure..

Exercise 10: Exploratory panel

Start with “Exploratory subgroup results; interpret with caution.”. Draft a target-language version or an English back-translation that an independent reviewer can understand. Then write two notes: which information must survive, and which tempting simplification would be wrong. The answer key is that keep exploratory qualifier and scope.. A typical wrong answer would writing confirms treatment benefit makes a claim not supported..

Quality-control checklist before publication

Source integrity: Confirm exact document, revision, date, provenance and intended audience. Semantic integrity: Verify actors, scope, uncertainty, quantity, conditions, negation and named categories. Structural integrity: Compare item count, numbering, headings, labels, codes, cross-references and display order. Reader integrity: Check how a representative reader interprets the statement. Review integrity: Keep a clear approval record and a way to correct mistakes.

Apply these checks twice: once before language polishing, when the original structure is still visible, and once after final layout. The second pass is essential because concise captions, mobile interfaces, extracted text, screen-reader order and page design can introduce a new mismatch after the language itself has been approved. For scientific figure captions and graph legends, choose a reviewer capable of recognising the domain-specific limits of each statement.

Frequently asked questions

Is literal translation the safest choice?

Not necessarily. A literal sentence can preserve word order and lose meaning. The safer approach is to keep facts, scope, qualifiers and reader consequences stable while allowing natural grammar in the target language. Use back-explanation and field alignment to test whether the intended information survived.

For scientific figure captions and graph legends, make the answer practical: use panel identifiers and reading order as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

What should a beginner check first?

Check the high-consequence parts first: negation, conditions, numbers, dates, actors, categories, uncertainty and reference identifiers. Once those are secure, improve rhythm and readability. This order prevents fluency editing from disguising a material meaning error.

For scientific figure captions and graph legends, make the answer practical: use axes and scales as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

Can machine translation produce the first draft?

It can be a starting aid, but the output still needs source comparison, domain review and reader testing. An automated system may not know which seemingly minor qualifier determines a real decision. Treat machine suggestions as unverified drafts and record human corrections.

For scientific figure captions and graph legends, make the answer practical: use variables and units as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

Should every technical term stay in English?

No. Use an accepted target-language term when one exists and accurately fits the source. Keep official names and identifiers where required. If a term has no established equivalent, define it in context or flag it for specialist decision rather than inventing an authoritative-sounding label.

For scientific figure captions and graph legends, make the answer practical: use legend marks and line styles as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

When is a translator’s note justified?

Use a clearly distinguished note when the source is genuinely ambiguous or an unfamiliar term needs explanation and the publishing brief permits notes. Do not use a note to conceal a guess. Separate the translator’s explanation from the source author’s claim.

For scientific figure captions and graph legends, make the answer practical: use sample size and independence as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

How can a reviewer catch an omitted condition?

Read the target text alone, identify who is included or excluded, then compare with the source. Try two boundary cases, one that should satisfy the condition and one that should not. If the translated version classifies them differently, revise it.

For scientific figure captions and graph legends, make the answer practical: use error-bar definitions as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

What is the difference between proofreading and accuracy review?

Proofreading checks target-language form, spelling, consistency and readability. Accuracy review also compares the source and target claims. A grammatical sentence may still convey the wrong evidence, measurement, limitation, condition or status. Both checks are needed.

For scientific figure captions and graph legends, make the answer practical: use confidence intervals as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

Does one glossary solve the entire problem?

No. Glossaries help with recurring terminology, but a correct term placed in the wrong relationship can still mislead. The document’s structure, qualifiers, version and referents must be reviewed independently. A glossary is part of the workflow, not the final quality certificate.

For scientific figure captions and graph legends, make the answer practical: use p-values and statistical tests as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

How do you check a translation without knowing both languages?

Use qualified bilingual and domain reviewers, structured alignment fields and a separate target-reader comprehension check. A monolingual reviewer can assess what the target communicates, but cannot alone certify that it equals the original source.

For scientific figure captions and graph legends, make the answer practical: use limits of detection as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

What if the original text contains an error?

Do not silently correct the official source by translating something else. Flag the discrepancy to the owner and request a corrected source or an approved explanatory note. Maintain a clear distinction between the original statement, an identified error and a later corrected edition.

For scientific figure captions and graph legends, make the answer practical: use logarithmic scales as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

How does this relate to search engine optimization?

Use natural query language where readers look for help, but do not sacrifice precise meaning to repeated keywords. Place the answer early, use informative headings and examples, and link to adjacent owners instead of rewriting their entire jobs. Search clarity follows editorial clarity.

For scientific figure captions and graph legends, make the answer practical: use association versus causation as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

How should teams handle updates?

Version the source and target together, compare the changed elements, ask whether linked sentences or visuals depend on them, and repeat approval on the final display. Do not assume a translation is still current merely because its URL remains unchanged.

For scientific figure captions and graph legends, make the answer practical: use reference to supplementary data as a sample checkpoint. A source-to-target review should end with an observable pass or fail, not just a feeling that the translated sentence reads well. If the reviewer cannot state the protected distinction clearly, mark the item unresolved and seek the appropriate expertise.

Sources, scope and further reading

These references support the subject distinctions and quality approach. They are not a substitute for the current official version of a standard, governing instrument, museum collection record or particular study. Check their applicability to the source’s jurisdiction and date, especially where later editions use different wording.

Your next useful route

For meaning and phrasing, visit the Vocabulary Learning Hub and the How English Works library. Those routes build vocabulary precision, sentence interpretation and the habits required to explain difficult information accurately. If numbers or reporting categories are involved, read how to translate decimals, measurements and percentages without changing quantities.

For documents where conditions or legal scope matter, the separate contract obligations and exceptions guide has a different job. When educational credential decisions are involved, use the academic transcript translation guide. For research participants and consent, use the clinical research consent guide. For technical metrics and limitations, see the AI model cards and benchmark translation guide. These links are intended as distinct routes rather than substitutes for the present page.

The central discipline is simple to state and demanding to practise: preserve the source’s meaning, make the target understandable, and never manufacture confidence that the original evidence does not provide. The best translation is one that readers in different languages can use to reach the same warranted conclusion. That is the standard to return to whenever a sentence looks polished but its consequences are uncertain.

Top Ways to Translate Correctly | Translate Scientific Figure Captions, Legends and Uncertainty Without Changing Results