Translation Review Services

What Our Translation Review Services Cover

Most of our work begins with source documents. This service begins with someone else's target text. Clients bring us translations from an incumbent agency, from a distributor who "handled it locally," from a bilingual employee, or from a machine translation pipeline, and they ask the same question in different words: can we trust this? Translation review services exist to answer that question with evidence rather than reassurance.

The service has four distinct shapes. A third-party review evaluates an existing translation against its source and produces a scored report. Technical editing goes further and corrects the text, sentence by sentence, until it meets publication standard. Proofreading, in the strict sense, is a monolingual pass over the target text alone, hunting errors of language and mechanics without consulting the source. Publication QA checks typeset proofs after desktop publishing, where a correct translation can still be wrecked by a bad line break or a swapped caption. Each is quoted separately, because each answers a different question at a different price.

Two things this service is not. It is not a certification: if you need a signed statement of accuracy for an official filing, that is a different deliverable, described on our certified technical translation page. And it is not the built-in second pass that already accompanies every translation we produce ourselves; that revision step is internal to our workflow. Review services are for text we did not write, which is precisely why clients trust the verdict.

The MQM Scorecard: How We Grade a Translation

Marked-up files alone invite arguments. A vendor can dismiss corrections as "preferential changes," and without a framework, the dispute goes nowhere. We grade against MQM (Multidimensional Quality Metrics), the open error typology developed under the EU's QTLaunchPad program and now the de facto standard for translation quality evaluation. Every error found is logged, categorized, and weighted by severity, which turns an opinion into a score.

Category	What it captures	Example from a real review
Accuracy	Mistranslation, omission, addition	"Normally closed valve" rendered as "valve closed under normal operation"
Terminology	Wrong or inconsistent technical terms	"Formwork" translated three ways in one bridge specification
Linguistic conventions	Grammar, spelling, punctuation	Decimal commas left in values destined for US readers
Fluency	Awkward or unclear phrasing	Instructions readers must parse twice to follow once
Style and locale	Register, units, date formats, audience fit	Metric-only dimensions in a manual for American field crews

Minor errors cost 1 point, major errors 5, and critical errors 10, normalized per 1,000 words. The report states the resulting score, the pass threshold agreed with you up front, and every logged error with its location, severity, and a suggested correction. You see exactly why the text scored as it did, and your vendor sees a case they can respond to line by line rather than a vague complaint.

Bilingual Review or Monolingual Proofreading: Choosing the Right Depth

The most common scoping mistake is buying the cheap option for a problem only the expensive one can find. Monolingual proofreading reads the target text alone. It catches spelling, grammar, punctuation, formatting slips, and sentences that make no sense. What it cannot catch, by definition, is a fluent sentence that says the wrong thing. A mistranslated pressure limit reads perfectly; only a reviewer comparing it against the source will spot it.

Bilingual review costs roughly twice as much per word because the reviewer processes two texts, not one. Our guidance is blunt: any content where an undetected meaning error carries safety, legal, or financial consequences gets bilingual review. That covers specifications, procedures, anything in a regulatory submission, and safety information of any kind. Monolingual proofreading is legitimate for content where the risk is reputational rather than physical: brochures, web copy, internal newsletters. When budget forces a choice on a large corpus, we bilingually review the high-risk sections and proofread the rest, and the quote shows the split explicitly.

Revise or Retranslate: Where the Economic Threshold Sits

Editing a bad translation can cost more than translating from scratch, and a vendor who never says so is selling you hours. Fixing a text means reading the source, reading the target, diagnosing the error, and writing the correction: four operations where fresh translation needs one. Our working threshold, calibrated over fifteen years of review projects, is that when more than roughly a third of segments need substantive correction, retranslation is cheaper, faster, and produces a better text. Below that line, editing wins. Near it, we sample further before recommending either path.

This is why every engagement starts with a diagnostic sample of 2,000–3,000 words drawn from different sections of the corpus. The sample review is inexpensive, takes two to three business days, and tells us the defect rate with enough confidence to project cost across the whole volume. Sometimes the news is good and we say so; roughly a quarter of the corpora we sample score well enough that we recommend leaving them alone apart from terminology harmonization. If the source of the problem is raw machine translation, the fix is a different discipline with its own method and pricing, covered by our machine translation post-editing services.

Publication QA: The Last Check Before Print

A translation can leave the linguist perfect and reach the reader broken. Desktop publishing introduces its own error class: text expansion pushing a callout out of its frame, a hyphenation engine breaking a chemical name illegally, Arabic reflowing left to right, a figure caption swapped with its neighbor, or an unconverted tolerance surviving in a table nobody re-checked. Publication QA is a structured pass over the typeset proof, in PDF or on paper, by a linguist reading the layout as the end reader will.

The checklist runs to 30 points and includes text integrity against the approved translation, hyphenation and line-break legality in the target language, completeness of headers, footers, and tables of contents, figure and caption alignment, and correct rendering of numbers, units, and warning symbols. Findings come back as annotated PDFs your layout team can execute directly. We recommend this pass on anything that goes to press or into a controlled document system; it typically costs 5–10% of the original translation and has repeatedly caught errors that would have forced a reprint of an entire manual run.

Vendor Audits Before You Sign, or Before You Renew

An increasing share of this work is procurement support. Before renewing a translation contract, or before handing a new vendor a critical program, clients commission a one-time quality audit of representative work. We score the samples on the MQM scorecard, benchmark them against the defect rates we see across comparable technical content, and give you a short written opinion you can put in front of management.

One engagement shows the pattern. A device manufacturer preparing an EU market expansion asked us to audit 60,000 words of translated IFUs and test protocols from their incumbent vendor before renewing a three-year contract. The stratified sample scored acceptably on accuracy but failed on terminology: the same sensor component carried four different names across the corpus, a defect their reviewers had each fixed locally without anyone seeing the whole. The client renewed the contract anyway, but with a termbase requirement and a quarterly audit clause written into it. That is the point of an audit: better decisions, not automatic vendor changes. Manufacturers in the medical device industry use these audits ahead of notified-body submissions, and firms in civil engineering use them before joint-venture bids where a partner supplied the translations.

Process, Deliverables, and Turnaround

Every review engagement follows the same five steps, compressed or expanded to fit the volume:

1. Scoping call. We agree on the corpus, the review depth per document class, the pass threshold, and whether you want corrections applied or findings only.
2. Diagnostic sample. 2,000–3,000 stratified words, scored and reported in 2–3 business days, with a firm quote for the remainder.
3. Full review. Reviewers work at 1,000–1,500 words per hour for bilingual review, roughly double for monolingual proofreading, logging every finding as they go.
4. Quality report. Scorecard, error log with locations and severities, corrected files if commissioned, and a one-page summary written for non-linguists.
5. Debrief. A 30-minute call to walk through the findings, including, when relevant, what to ask your current vendor to change.

Reviewer independence is structural, not incidental. The person who scores a text never has a stake in retranslating it: our review team is compensated on review work, and our ISO 17100-compliant process keeps evaluation and production roles separate on every engagement. Assessments are performed by the same caliber of specialist described on our technical translator page, working under ATA and GALA professional frameworks, from our base in Texas with coverage across US time zones.

What a Review Costs, and How We Bill It

Bilingual review and technical editing are billed per source word, at roughly 50–60% of full translation rates, with the exact figure set by language pair, subject matter, and the defect density found in the sample; heavily flawed text takes longer to fix and the quote says so openly. Monolingual proofreading bills at about half the bilingual rate. Publication QA and vendor audits are billed hourly with a capped estimate, since page count and finding density drive the effort more than word count does. Every quote itemizes these components so you can drop one without renegotiating the rest.

Where does the money come back? Clients who run a diagnostic before a large program avoid paying editing rates for text that needed retranslation, or retranslation rates for text that needed a light edit. One engineering firm brought us translated engineering reports from three regional offices, convinced the whole set needed to be redone; sampling showed two offices' work was sound, and the third's was not. They paid for one-third of the retranslation they had budgeted. If you are weighing a similar decision, this page lists the neighboring services a review often leads into, and the quote form takes samples in any format, including scanned proofs.