If you are searching for how to translate right-to-left languages, RTL translation, bidirectional text translation, or how to handle Arabic, Hebrew, Persian, Urdu and other right-to-left writing systems inside websites, documents or apps, translation accuracy is only half the job. The words can be correct while numbers, punctuation, brackets, URLs or product codes appear in the wrong visual relationship. A target reader may then misread a date, copy the wrong identifier, follow a reversed step sequence or assume a symbol belongs to the wrong phrase.
Right-to-left translation becomes especially demanding when a line contains both RTL and LTR material. This is called bidirectional text: different parts of the same line follow different writing directions. Latin brand names, email addresses, web links, model numbers, mathematical expressions and digits can sit inside a right-to-left sentence. Good translation must preserve meaning, but good localization must also preserve grouping, reading order and usability after the text is rendered.
This guide gives a practical system for translating RTL and mixed-direction content without breaking names, numbers, punctuation, navigation or interface meaning. It focuses on the translator’s decisions: what should be translated, what must remain protected, how punctuation can become ambiguous, when layout testing is required, and how to verify the final rendered page instead of trusting a raw text field.
Direction is part of the reading experience
Writing direction is not the same as semantic order. A language may be read primarily from right to left while individual elements inside it retain a left-to-right order. That is why a translator should think in directional runs rather than assuming an entire line behaves uniformly. A phone number, URL or chemical formula may need to remain internally stable even when the surrounding sentence reads in the opposite direction.
The main risk is false confidence. In an editor, a string can look plausible because software is actively reordering characters for display. When copied into another system, the same sequence may render differently. Conversely, a raw string that looks strange to someone unfamiliar with RTL may be correct once displayed by a proper engine. Translation review must therefore happen in the target environment whenever the content is operational rather than purely literary.
Another risk is confusing layout mirroring with language translation. Interfaces often mirror navigation, alignment and spatial relationships for RTL languages, but not every object should flip. A play icon, brand mark, chart, map, progress bar or physical-direction arrow may have its own meaning. The translator should flag these issues but avoid changing them blindly as if all visual orientation were linguistic.
The core principle is simple: preserve the semantic relationships first, protect data-like strings second, and test visual rendering third. Those three checks catch most serious bidirectional failures.
A reliable RTL and bidirectional translation workflow
1. Mark protected strings before translation
Identify URLs, email addresses, usernames, file paths, product codes, version numbers, formulas and other strings that should normally remain exact. Protecting them before drafting prevents accidental reordering, character substitution or translation.
2. Separate directional runs
Read the line as groups: RTL words, LTR names, digits, punctuation and symbols. This helps you see where visual ambiguity may occur and where a neutral character such as a slash, dash or bracket sits between runs.
3. Translate meaning without hand-reversing text
Write target-language grammar normally. Do not manually reverse word or character order to make a raw editor look right. Proper text systems apply directional rules during rendering.
4. Check punctuation in context
Punctuation often sits at boundaries between directional runs. Verify brackets, colons, slashes, dashes, percent signs and quotation marks in the rendered output so readers can tell which phrase or number they belong to.
5. Preserve internal order of data
Model numbers, dates, times, decimal values, identifiers and URLs may have conventions independent of paragraph direction. Translate labels around them while keeping the validated data itself stable.
6. Test the real interface or document
Preview the final web page, app, PDF, slide or form. Line wrapping, alignment and component mirroring can introduce problems invisible in a translation memory table or spreadsheet.
7. Use a native-reader QA pass
A fluent target-language reader can detect grouping problems that a source-language reviewer may overlook. Ask them to perform the task the text supports: copy a code, follow a step, enter a number or navigate a menu.
8. Record recurring RTL rules
If a project repeatedly mixes the same kinds of strings, document conventions for numbers, Latin product names, punctuation, placeholders and mirrored interface components. A stable project rule prevents every translator from improvising a different solution.
The bidirectional text problems that cause real errors
1. Latin brand names inside RTL sentences
This issue appears when a brand or product name remains in Latin script inside translated prose. A source unit such as an Arabic or Hebrew sentence containing a Latin product name can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Keep the official brand form unless an authorised local form exists, and test spacing and punctuation around it so the brand does not visually attach to the wrong word. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the brand remain readable in its official order and clearly belong to the intended surrounding phrase? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
2. URLs
This issue appears when a web address appears inside an RTL paragraph. A source unit such as https://example.com/help can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Do not translate the URL unless the site provides a deliberate localised path. Treat it as protected data and make sure punctuation around it does not become visually confusing. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, can a user copy, click or type the address correctly from the rendered page? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
3. Email addresses
This issue appears when an email address contains LTR characters inside an RTL line. A source unit such as support@example.com can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Keep the address exact and avoid inserting translated spaces or punctuation into it. Translate only the surrounding instruction. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does copy-paste reproduce the exact address and does the visual line make the address boundaries obvious? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
4. Phone numbers
This issue appears when contact information is displayed inside an RTL form or sentence. A source unit such as +65 6123 4567 can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Preserve the validated number and country code order according to the project’s locale convention. Do not reverse digits to match paragraph direction. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, can a reader dial or copy the number without ambiguity? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
5. Dates
This issue appears when the target language has local date conventions while digits remain directional units. A source unit such as 18/09/2026 can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Localise the date format only when the project calls for it, and make month/day/year interpretation explicit where ambiguity is possible. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the rendered date preserve the correct calendar meaning and separator grouping? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
6. Times
This issue appears when a line combines time values, punctuation and translated labels. A source unit such as 14:30 UTC can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Preserve the time value and zone unless conversion is explicitly required. Translate labels around it while checking colon placement and visual grouping. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, can the reader identify hours, minutes and time zone correctly? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
7. Percentages and decimals
This issue appears when numbers combine digits with percent signs, decimal separators or units. A source unit such as 12.5% growth can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Follow target-locale numeric conventions where required but preserve the numerical value exactly. Be careful that percent signs and minus signs remain associated with the intended number. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the target display the same magnitude, sign and unit? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
8. Currency values
This issue appears when currency symbols, codes and amounts sit within RTL prose. A source unit such as USD 1,250.00 can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Follow the project’s currency style and local numeric convention without changing the amount or currency identity. Do not infer conversion unless explicitly requested. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, is it unambiguous which symbol or code belongs to which value? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
9. Parentheses
This issue appears when an RTL phrase contains an LTR term or acronym inside brackets. A source unit such as translated term (API) can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Check the rendered brackets rather than manually swapping them. Ensure the parenthetical item remains attached to the correct phrase. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, do opening and closing brackets visually surround the intended content in the final environment? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
10. Colons and labels
This issue appears when forms display translated field labels followed by LTR data. A source unit such as Email: support@example.com can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Translate the label and preserve the data. Test the colon and spacing so readers see a label-value pair rather than a scrambled run. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, is the relationship between label and value visually obvious? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
11. Bulleted and numbered lists
This issue appears when an RTL document contains ordered steps and embedded LTR strings. A source unit such as 1. Open URL 2. Enter code can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Keep the logical step order and let the target layout system place list markers appropriately. Do not reverse sequence numbers. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the reader perform steps in the same intended order? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
12. Arrows and directional icons
This issue appears when instructions use arrows whose meaning may be spatial rather than linguistic. A source unit such as Click → to continue can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Determine whether the arrow represents reading progression, physical direction, playback or a fixed control. Mirror only when the interface meaning requires it. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, would mirroring the icon change the action or physical direction being communicated? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
13. Breadcrumbs and navigation paths
This issue appears when a website shows hierarchical paths alongside RTL labels. A source unit such as Home > Help > Translation can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Localise labels and follow the platform’s RTL breadcrumb convention while preserving hierarchy. The sequence must still represent parent-to-child navigation. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, can users tell which page is broader and which is current? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
14. Model numbers and SKUs
This issue appears when products have fixed alphanumeric identifiers. A source unit such as XR-500B can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Treat identifiers as protected strings. Translate the product description, not the code. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the final text preserve every character, hyphen and position in the identifier? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
15. Version numbers
This issue appears when software or documents reference versions inside translated text. A source unit such as v2.4.1 can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Keep the version string exact unless the product itself localises its prefix. Do not reinterpret dots as decimal punctuation. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the target point to the same release? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
16. Mathematical expressions
This issue appears when an RTL explanation contains equations or operators with their own conventional order. A source unit such as x + 3 = 7 can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Preserve mathematical notation and translate the surrounding explanation. Do not reverse an equation to match paragraph flow. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the expression still state the same mathematical relationship? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
17. Source code and commands
This issue appears when documentation embeds commands, function names or code tokens. A source unit such as npm run build can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Protect code unless the programming language explicitly allows localisation of comments or user-facing strings. Translate the explanation, not executable syntax. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, can the copied command still run or identify the same function? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
18. Placeholders and variables
This issue appears when a translated string contains tokens that software replaces at runtime. A source unit such as Hello, {user_name} can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Keep the placeholder exact and position it where target grammar requires if the localization system supports reordering. Never translate the variable name unless specified by the system. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, does the runtime still substitute the intended value and does the sentence remain grammatical for likely values? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
19. Tables with mixed scripts
This issue appears when rows and columns contain RTL labels alongside LTR codes and numbers. A source unit such as a pricing table with Arabic labels and Latin SKUs can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Translate labels and preserve data columns, then test column alignment and reading order in the final table layout. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, can readers associate each code and value with the correct row and heading? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
20. PDFs and exported documents
This issue appears when content that looked correct in an editor is exported to PDF. A source unit such as an RTL handbook with mixed Latin references can be semantically correct and still display in a way that makes the target reader misread sequence, grouping or ownership. The translator therefore needs to distinguish language direction from the direction of embedded items. A right-to-left paragraph can contain left-to-right numbers, URLs, model codes, Latin names and formulas; a left-to-right paragraph can contain a right-to-left quotation or name.
Always inspect the export. Font substitution, line wrapping and text shaping can reveal issues that were hidden in the authoring environment. Do not repair a display problem by changing the meaning-bearing order of words unless grammar actually requires it. Many bidirectional problems come from formatting, punctuation or software handling rather than translation. The safest workflow preserves the validated text, then tests it in the real target environment where the browser, app, document engine or publishing system will apply bidirectional layout.
For quality assurance, are lines, punctuation, page numbers, bookmarks and copied text all still usable after export? Inspect both visual order and logical meaning. Ask a target-language reader to verify the rendered output, not merely the raw string. Copy-paste tests can also reveal whether a line that looks correct is stored in a confusing order. The final check is practical: can the reader identify what belongs together, enter the data correctly, follow the arrow or step sequence, and understand the same relationships as the source reader?
Common failure modes
1. Manually reversing characters
This usually creates corrupted logical text that may display differently elsewhere. Write the language normally and let standards-compliant software handle direction.
2. Reversing digits
Paragraph direction does not mean numbers should be typed backwards. Preserve validated numeric values and test their rendered grouping.
3. Translating protected data
URLs, emails, codes, variables and model numbers are often data rather than prose. Changing them can make the content unusable.
4. Trusting the CAT-tool grid
A translation table may not reproduce the final web or app layout. Bidirectional QA must include the actual rendering environment.
5. Mirroring every icon
Some icons represent physical direction, media controls or brand identity rather than reading progression. Mirror only when meaning supports it.
6. Ignoring punctuation boundaries
Neutral punctuation can appear visually attached to the wrong directional run. Check rendered parentheses, colons, slashes and dashes carefully.
7. Letting LTR names break grammar
A retained Latin name still sits inside target-language syntax. Place it where grammar requires and check spacing around it.
8. Reviewing only visually
A line can look correct while copy-paste order is wrong, or look unfamiliar while being logically correct. Check both visual rendering and underlying usability.
Worked practice
Practice 1: Email in a support sentence
Situation: An RTL sentence tells users to contact support@example.com. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Translate the instruction, keep the address exact, then preview the rendered line and copy-paste the address to verify boundaries and stored order. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 2: Price and currency
Situation: A product page shows USD 149.99 inside RTL text. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Keep currency identity and amount exact, localise formatting only according to project rules, and verify that the symbol or code still attaches to the intended value. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 3: Numbered instructions
Situation: Three translated steps each contain an LTR command. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Preserve step sequence and protected commands; use the RTL list layout without reversing numbers or executable tokens. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 4: A breadcrumb
Situation: Home > Account > Security is localised for an RTL website. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Translate labels, preserve hierarchy, and let the target interface render separators according to its RTL design rather than manually swapping characters. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 5: A placeholder
Situation: The source string is “Welcome, {first_name}”. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Keep {first_name} exact, position it according to target grammar if the platform supports placeholder reordering, and test with multiple sample names. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 6: A mixed-script heading
Situation: A target-language heading contains the Latin acronym API. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Keep the recognised acronym, check spacing and brackets, and make sure it visually belongs to the correct noun phrase. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 7: A phone number
Situation: A form displays an international phone number beneath an RTL label. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Translate the label, preserve the number’s validated order and country code, and test dialing/copying from the final interface. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
Practice 8: An exported PDF
Situation: The web preview is correct but the final deliverable is PDF. First mark every directional run: target-language words, Latin text, digits, punctuation, symbols and protected strings. Then decide which parts are translated and which parts must remain exactly as data.
Reasoning: Inspect the PDF itself, including copy-paste, page headers, bookmarks, line wraps and fonts. Rendering is part of translation QA. The correct answer must be tested in context because raw source editors, spreadsheets, browsers and mobile interfaces can render the same character sequence differently. Translation review should therefore include a visual pass and a semantic pass.
AI, CAT tools and bidirectional text
CAT tools can help protect tags and placeholders, but the segment view is not a substitute for a rendered preview. Use preview modes, screenshots or exported builds whenever the project includes mixed-direction strings. Mark protected content so translators do not “fix” it into broken data.
AI translation can generate fluent RTL prose, but it may not know the final interface constraints and can alter protected strings unless instructed not to. Give explicit rules for URLs, variables, codes, numbers and names, then verify them programmatically or with a concordance check.
Browser developer tools, pseudolocalization, screenshots and native-reader testing are valuable because bidirectional bugs often appear only after layout. A translation workflow is stronger when linguistic review and functional testing are separate gates rather than one combined glance.
Connections to the wider eduKate translation system
Bidirectional text makes the larger translation principle visible: meaning is carried not only by vocabulary but also by structure, reference and presentation. The general method is developed in Master Art of Translation | The Complete System for Moving Meaning Between Languages. Language knowledge connects to the Vocabulary Learning Hub, while sentence structure, reference and information order connect to How English Works. A successful RTL translation preserves all of those layers and then proves that the rendered page still lets the reader use them.
FAQ
Do right-to-left languages reverse numbers?
No general rule says to reverse digit strings simply because the paragraph is RTL. Preserve the validated numeric value and follow target-locale conventions for formatting.
Should URLs be translated?
Usually keep the actual URL unless the website has an intentional localised path. Translate the link text around it when appropriate.
Why do brackets look wrong in mixed text?
Brackets and punctuation sit near directional boundaries and can render unexpectedly depending on surrounding characters. Always test the final display.
Can I manually reverse RTL text to make it look right?
No. Manual reversal usually damages logical order and portability. Use proper RTL-capable software and styles.
Should interface arrows always flip in RTL?
No. Flip only when the arrow represents reading or navigation direction. Physical-direction, media or brand-specific icons may need to stay unchanged.
How should placeholders be handled?
Keep placeholder syntax exact. Reorder the placeholder within the sentence only if the localization system supports it and target grammar requires it.
What is bidirectional text?
It is text that contains runs with different writing directions, such as an Arabic sentence containing a Latin URL or product code.
Is a linguistic review enough for RTL translation?
No. Functional rendering review is essential for websites, apps, forms and exported documents because display can change relationships among strings.
How do I test an RTL translation?
Preview it in the final environment, inspect punctuation and mixed strings, copy protected data, complete representative tasks and get a fluent target reader to review it.
Does RTL translation affect SEO?
The semantic SEO principles are the same, but page templates, titles, breadcrumbs and interface components must render correctly so users and crawlers receive coherent content.
Final checklist
- Have I marked URLs, emails, codes, variables and identifiers as protected?
- Have I separated RTL and LTR directional runs mentally?
- Did I avoid manual character reversal?
- Are numbers, dates, times and currency values unchanged in meaning?
- Are brackets, colons, slashes and dashes visually attached to the right content?
- Do arrows and icons preserve their actual function?
- Have I previewed the text in the real website, app, PDF or document?
- Can protected strings be copied correctly?
- Has a fluent target-language reader checked the rendered result?
- Have I tested representative tasks, not just read the page?
Right-to-left translation is a strong reminder that language does not live only in dictionaries. Readers encounter meaning through lines, screens, fields, numbers, links, symbols and actions. The safest method is to protect data, translate language normally, let proper software handle direction, and then test the rendered result as a user would experience it. When semantic accuracy and visual usability are both verified, mixed-direction text stops being a fragile edge case and becomes a manageable part of professional translation.