Even after manual translation and technical document translation are finished, a multilingual manual still needs checking before it goes to DTP. The automated QA built into CAT tools reliably catches structured errors such as untranslated segments, missing tags, and number mismatches. It is far less reliable on the defects that actually stop a layout: text baked into images, hidden or locked layers, corrupted glyphs, lost emphasis, mistranslated product names, non-standard safety signal words, and broken cross-references, URLs, and variables. Manual translation QA has to run as a gate of its own.
The gap widens when AI and machine translation move into the production workflow. MT output reads fluently at the sentence level. Whether that file can be typeset, published, printed, and archived is a separate question.
What is a translation integrity check?
A translation integrity check is a pre-DTP review that confirms a translated manual file is ready for production. It verifies that images, hidden layers, character rendering, emphasis, product names, safety signal words, cross-references, URLs, and variables are all correct before layout work starts. Within a manual localization workflow it functions as a process gate. Linguistic quality scoring is a separate activity with a separate purpose.
- Who this is for Technical writers, manual owners, localization managers, DTP and QA staff, and manufacturer documentation teams running AI translation in house.
- When to use it On receipt of translated files, before multilingual DTP starts, when reviewing AI or MT output, and whenever you need to cut rework in the final week before delivery.
- Use with The companion Translation Integrity Checklist, which turns these 15 checks into a sign-off form.
This guide covers general checks that apply across industries and product types. Product-specific regulatory wording, safety marking requirements, customer brand guidelines, and market-specific legal obligations require separate review. It does not replace a standard, a regulation, or a customer style guide.
Why check translation integrity before DTP?
In technical manual production and manual localization, translation and DTP look like separate stages. In practice they are tightly coupled. When defects survive into the translated file, the DTP operator cannot lay out structure, tables, images, captions, variables, and hyperlinks with any confidence. If the same defect surfaces later, during PDF or HTML output, the fix is rarely a quick edit. It escalates into re-translation, re-layout, and a second full review.
A translation integrity check is therefore more than a language quality assessment. It is the decision point that answers one question: can this file move forward into editing, multi-channel output, print, and archiving?
| Translated files received | Integrity check | DTP | Final output | Launch / delivery |
|---|---|---|---|---|
| AI, MT, or vendor translation arrives | Images, glyphs, terminology, links, variables verified | Typesetting and layout applied | PDF, HTML, print files generated | Market release or customer delivery |
Core principle. The later a defect is found, the more expensive it is to fix and the more schedule risk it carries. Grade A items must be at zero before DTP begins.
What CAT tool QA catches, and what it misses
Trados and comparable CAT platforms are strong at anything that can be defined mechanically: untranslated segments, missing tags, number mismatches, terminology violations. Output quality in a multilingual manual is not decided at segment level. Images, layers, fonts, UI labels, safety marking, cross-references, regional links, and variable values all belong to the document production process, and that is where they fail.
| Area | Automated QA handles this well | Needs a separate pre-check |
|---|---|---|
| Text | Untranslated segments, number mismatches, missing tags | Deciding what must be translated and what must stay in English. AI renderings that read fluently but are wrong in context |
| Formatting | Some tag and style errors | Placement of bold, color, underline, and italics. Per-language conventions for emphasis |
| Non-text elements | Limited detection | Text inside images, charts, and callouts. Hidden layers and locked text boxes |
| Publishing elements | Limited detection | Cross-references, TOC links, regional URLs, header and footer variables, version data |
| Risk exposure | Segment-level errors | DTP rework, reprint, launch delay, customer claims, regulatory and product liability exposure |
Integrity defects that recur in AI and MT output
AI and machine translation are fast and often fluent. Neither one guarantees the production conditions a manual has to satisfy. The three defects below are typical: the sentence itself may be acceptable, while the file is not publishable.
| Defect | What it looks like |
|---|---|
| Image dropped | Source: After sending an SOS message, the icon [icon graphic] will display on the watch screen. Target: The inline icon is gone. The translator replaced the graphic with a text description, so the reader cannot match the instruction to the screen. |
| Corrupted glyphs | Source: Quick launch: Set the device to launch the camera by pressing the Home key twice quickly. Target: The traditional Chinese string renders as boxes and question marks because the font in the layout file has no glyph coverage for the script. |
| Emphasis lost | Source: To view the video, tap the preview thumbnail and tap PLAY MOTION PHOTO. Target: In the Arabic file the bold UI label is rendered as plain body text. The user can no longer tell which words name the on-screen control. |
The four categories of the 15 checks
These checks are deliberately industry-neutral. They apply to consumer electronics, automotive, industrial and construction equipment, medical devices, and robotics alike, because every one of these programs publishes technical documentation and user manuals in multiple languages. There are 15 checks across four categories.
| Category | What it verifies before DTP | Checks |
|---|---|---|
| A. Translation completeness and integrity | Text inside images, hidden and locked elements, glyph rendering, emphasis formatting | 5 |
| B. Terminology and UI consistency | Product and model names, brand names, molded and engraved product text, safety signal words, trademark symbols | 4 |
| C. Numbers, punctuation, capitalization | Per-language punctuation and quotation rules, capitalization standards, native numeral systems | 3 |
| D. Cross-references, links, variables | Cross-reference numbering, hyperlinks, regional URLs, header and footer variables | 3 |
Severity grades and the sign-off rule
The 15 checks do not carry equal weight. Grade them by the damage they do downstream, then enforce one hard rule: Grade A must be at zero before DTP starts.
| Grade | Name | Definition and consequence if missed |
|---|---|---|
| A | Critical / mandatory | A single miss can trigger a launch delay, a recall, a reprint, or a legal dispute. Must be zero at sign-off. |
| B | Major / recommended | A miss leads to customer claims, rework, and erosion of brand trust. Target zero wherever possible. |
| C | Minor / supporting | Low impact individually. Accumulated across languages and releases, these drive quality variance and downstream operating cost. |
The 15 checks in detail
The table below expands each checklist line into working guidance. In production, record every item twice: once by the person who received the files (Self), once by an independent reviewer (Reviewer).
| No | Check and acceptance criterion | Grade |
|---|---|---|
| 1 | Is text inside images and charts translated? Check screen captures, callouts, chart labels, and warning graphics. Any English string that must stay in English is an approved exception and gets logged as one. | A |
| 2 | Are indexes, locked layers, hidden layers, and text boxes translated? Content that CAT export skips or that the source file hides is the single most common source of late-stage content loss. Compare object counts before and after extraction. | A |
| 3 | Are there any corrupted characters or missing glyphs? Apply the correct font for each script, then verify in the translated file, the layout file, and the exported PDF. CJK, Arabic, Indic scripts, Thai, and Vietnamese fail most often. | A |
| 4 | Is in-line emphasis such as bold and color preserved correctly? Compare emphasis position against the source. Safety text, UI labels, and named controls must keep their emphasis or the reader loses the cue. | B |
| 5 | Are italics and underline applied according to target-language convention? Simplified and traditional Chinese, Japanese, and RTL languages restrict or avoid italics and underline. Remove the formatting where the convention calls for it. | B |
| 6 | Are product names, model numbers, and brand names handled per standard? Check against the brand guideline, the approved product name list, and the do-not-translate list. Machine translation localizes brand names readily. | A |
| 7 | Is molded, engraved, and nameplate text handled per standard? Text physically present on the product, on buttons, panels, and labels, must match the manual exactly. A mismatch here makes the instruction unusable. | A |
| 8 | Do safety signal words and safety statements follow the approved standard? Verify DANGER, WARNING, CAUTION, and NOTICE against ANSI Z535.4 and Z535.6, ISO 3864-2, and the customer’s approved warning library. Free translation of a signal word creates liability exposure that reaches well beyond style. | A |
| 9 | Are trademark symbols applied per the marking standard? Confirm that ™, ®, and © are retained, correctly positioned, and not silently dropped by the authoring or DTP software. | B |
| 10 | Do punctuation, quotation marks, signs, and units follow target-language rules? Check quotation styles, decimal separators, thousands separators, unit spacing, and plus and minus sign conventions per language. | B |
| 11 | Does capitalization follow target-language and customer style? English title case and sentence case do not transfer mechanically. Verify against the customer style guide and the conventions of each language. | B |
| 12 | Are native numerals applied consistently where required? Arabic, Persian, and comparable markets may require native numeral forms. Confirm the customer or market standard, then apply it consistently across the document. | C |
| 13 | Are cross-references, hyperlinks, and TOC links valid? Page numbers, section numbers, table and figure numbers, bookmarks, and TOC links must resolve in the final output rather than surfacing as error codes. | A |
| 14 | Are URLs valid and regional URLs correctly mapped? Test live links, country domains, local support pages, and QR destinations. A US manual pointing at a Korean support page is a support-cost problem on day one. | A |
| 15 | Are header, footer, language bar, and version variables preserved? Variable names, language codes, document version, and release data must survive translation unchanged and must match the release record. | B |
Five defects most often found just before delivery
Hansem Global treats the five defects below as first-line blocking items in its LQA and pre-delivery inspection process. If you have no dedicated LQA system, or if you run AI translation directly, these are the ones to clear before DTP.
| Defect type | What happens if it ships | How to block it before DTP |
|---|---|---|
| Untranslated text inside images and charts | The source language survives into layout. The result is re-translation plus image rework, usually discovered after the DTP budget is spent. | Export image, chart, screen capture, and callout text as a separate list and check it line by line. |
| Lost text boxes and hidden layers | Core content goes missing and the loss surfaces only after editing is complete, which puts the schedule at risk. | Compare object counts, hidden layers, and locked text before and after extraction. |
| Broken cross-reference variables | Page numbers, section links, and table or figure references appear as error codes in the final PDF or HTML. | Confirm variables survive the round trip, then run a link test on the generated output. |
| Free translation of safety statements | The wording departs from ANSI Z535, ISO 3864, or the customer’s approved library. In the US market this is failure-to-warn exposure, and it is the defect with the largest downside. | Reconcile every signal word and warning against the approved safety glossary and warning database. |
| Unmanaged translation of product names and trademarks | The global brand guideline is broken, the manual no longer matches the product UI or the physical label, and the customer files a claim. | Apply the do-not-translate list, the brand glossary, and the model name list before translation starts. |
Error or acceptable? Practical judgment calls
| Text inside images | ✗ Error | ✓ Correct |
|---|---|---|
| Body copy is fully translated, but button names, on-screen text, and warning labels inside the graphics are still in the source language. | Image, chart, and callout text is in scope for translation, and DTP verifies that no source-language string survives in the placed graphics. |
| Emphasis formatting | ✗ Error | ✓ Correct |
|---|---|---|
| The source puts a control name or UI label in bold. The target drops the bold, and the reader can no longer identify what to press. | Emphasis that carries meaning is preserved in the same position as the source. Where a language has a different convention, the formatting is adapted to the localization standard. |
| Cross-references and URLs | ✗ Error | ✓ Correct |
|---|---|---|
| “See page XX”, TOC links, QR destinations, and country support URLs are broken and do not resolve in the final output. | Internal links, external URLs, and regional domain mappings are verified before publication, and a click test is run on the final PDF and HTML. |
Running it as a two-stage cross-check
Batching the integrity check into the final week before launch defeats its purpose. The later a defect appears, the wider the correction radius and the more re-editing, re-review, and re-approval it demands. Split the work across two points: on receipt of the translated files, and immediately before DTP.
| Stage | Who performs it | What they do | Sign-off criterion |
|---|---|---|---|
| Stage 1: producer self-check | The person receiving the files, the project manager, or the translation reviewer | Walk all 15 checks with the checklist and clear every Grade A defect first. | Grade A at zero. Grade B and C findings logged with impact and a fix plan. |
| Stage 2: independent cross-check | A reviewer or QA lead who did not work on the translation | Re-verify the items that drive downstream cost: images, glyphs, product names, safety statements, URLs, variables. | Approve DTP start, or reject. A rejected file is corrected and re-verified. |
Dividing the work between automation and people
Send the repeatable, rule-shaped judgments to automated tooling first. Reserve human reviewers for the areas that require judgment: contextual flow, cultural fit, conformity with market-specific regulatory standards, and each customer’s brand rules. Hansem Global operates this division of labor as a pre-delivery inspection process built on LQA and its QA Suite of more than 60 automated checks.
How manufacturer documentation teams adopt manual localization QA
You do not need a dedicated LQA platform to reduce intake risk. Run the sequence below.
- Before translation starts, assemble the product name list, model numbers, brand names, UI strings, approved safety statements, and the do-not-translate list.
- If you use AI or MT, run the 15-point self-check immediately after translation, while the file is still cheap to fix.
- Drive Grade A items to zero before DTP. Anything you cannot close gets an exception reason and a named owner.
- Treat text inside images, charts, and callouts, plus hidden layers and locked text boxes, as full-population checks. Sampling does not find these.
- Before DTP begins, have a reviewer who did not touch the translation cross-check the critical items.
- Keep the completed checklist. It becomes the evidence trail for the DTP, regulatory, UI matching, accessibility, print, and data management stages that follow.
Frequently asked questions
- If our translation vendor has already run CAT tool QA, does the manufacturer still need to check? Yes. Vendor CAT QA is useful for untranslated segments, tags, numbers, and terminology. Only the manufacturer can confirm that text inside images, product and model names, molded product text, safety statements, regional URLs, and document version data match the actual product and the actual launch configuration.
- Does using AI translation make this check more important? It does. AI produces fluent sentences. It does not guarantee that product names are preserved, that safety statements follow the approved standard, that variable values survive, that text inside images is captured, or that per-language formatting conventions are applied.
- Who should own the translation integrity check? The first pass belongs to whoever receives the translated files, typically the project owner. The second pass belongs to a reviewer who was not involved in producing the translation.
- Can the same person do both the self-check and the reviewer check? It is not advisable. The point of a cross-check is to bring a second set of eyes to what the producer could not see. Wherever staffing allows, assign the reviewer check to someone with no involvement in the translation or the layout prep.
- How does this relate to ANSI Z535 and IEC/IEEE 82079-1? Those standards define what a compliant warning and a compliant instruction have to contain. This check confirms that the compliant content actually survived translation and file handling intact. A perfectly drafted warning provides no protection if the translated file drops it, corrupts it, or paraphrases the signal word.
- What is the difference between manual translation and manual localization? Manual translation converts the source sentences of a user manual or technical document into the target language. Manual localization extends that work to text inside images, safety marking, units and measurement conventions, regional URLs, and per-language formatting, so the document meets the requirements of the target market. The 15 checks in this guide confirm that a translated file has met those localization requirements before DTP begins.
Pre-Launch QA Series for Multilingual Manuals (8 parts)
The series follows a multilingual manual from translation intake to market release, one gate at a time. Each part pairs a guide with a self-check checklist.
- Part 1 · Translation integrity Catching defects in translated files before editing (this guide)
- Part 2 · DTP Multilingual typesetting and layout integrity
- Part 3 · Regulatory Market-specific regulatory and standards marking
- Part 4 · UI matching Synchronizing product screens with the document
- Part 5 · Accessibility EAA, WCAG, PDF/UA, and Section 508 conformance
- Part 6 · Multi-channel output Consistency across PDF, HTML, and mobile
- Part 7 · Print Bleed, CMYK, font embedding, and print readiness
- Part 8 · Data management Turning source files and TM into reusable assets