Answer
Translation Management System
A translation management system routes files, users, deadlines, and TM/MT around certified translation production — here is what a TMS owns, when it is
Agencies running multi-locale production ask one question before they buy anything: where does the spreadsheet stop working, and what takes over. A translation management system is the layer that takes over — routing files, users, and deadlines around the translation memory and machine translation engines the humans actually edit against. This page is written for translators and agency ops leads on the production side. It assumes a certified-lane stack with an ISO 17100-aligned LQA gate and a named-translator attestation at the end.
Key points
- A translation management system is the software layer that routes files, users, and deadlines around translation memory and machine translation, and stores the audit trail certified lanes reference.
- A TMS becomes necessary once a project exceeds two or three languages or involves multiple concurrent translators. Three or more locales pays the TMS back inside a quarter. or more locales pays the TMS back inside a quarter — and the ROI compounds as repeated segments draw on a growing TM corpus instead of being re-translated from scratch.
- A TMS alone does not certify a page, run an LQA scorecard, or produce a named-translator attestation — the human translator signs the certification, and the TMS exists to log every segment touch that makes that signature auditable.
- The LQA gate the TMS triggers prices at $40–$90 per 1,000 words against an MQM-style scorecard aligned to ISO 17100, and is mandatory on every certified lane regardless of language pair or document type.
- Governance rule: one TM per client per locale with lane-level metadata that prevents a certified-document segment from overwriting its marketing variant — without this separation, a legally reviewed phrase is silently replaced by a promotional equivalent during the next project run.
- A production-grade SaaS localization program has six distinct components. These are the TMS, CAT editor, QA tooling, AI first-draft engine, TM store, and connector layer.ngine, glossary and TM, and a release-trailing CI/CD workflow — each owned by a named role on the agency side, so the TMS is the orchestration hub rather than the full stack.
- Language variety adds another dimension to TMS governance: a single target language such as Spanish requires multiple locale-specific TMs covering distinct regional variants, each maintained separately to prevent cross-contamination of approved terminology.
What is a translation management system, and what does the TMS actually own?
A translation management system is the software layer that routes files, users, and deadlines around translation memory and machine translation. It owns workflow routing, vendor orchestration, engine selection, TM and term base storage, and reporting — the system of record every other tool reads from. Most agencies find it through the parent cluster on How Translation Agencies Win Clients and Manage Freelancers, where the TMS sits as the throughput backbone.
The six functions the TMS owns, from the production side:
- Workflow routing — moves a file from intake to pre-translation to human edit to LQA to delivery.
- Vendor and reviewer orchestration — assigns jobs to named translators and reviewers with the right credentials.
- Engine selection — picks the configured MT engine per locale and per content type.
- TM and term base storage — holds the segment pairs and glossary that every job reads from.
- Connector sync — subscribes to repositories and CMSes, extracts strings, pushes translated keys back.
- Program-level reporting — tracks volume, cost, leverage, and cycle time across clients and locales.
For the full stack walkthrough across platforms, connectors, and reviewer roles, see Software Translation Management: How Translators and Agencies Run a TMS Stack.
How does a TMS differ from a TM, an MT engine, and a CAT tool?
A TMS routes jobs and owns the audit trail; a TM stores approved segment matches; an MT engine generates first-draft output; and a CAT tool is the editor where the human post-edits. Confusing the four is the single most common buying mistake translators report.
| Layer | What it is | When you need it |
|---|---|---|
| Translation memory (TM) | Database of human-approved segment pairs that replays matches on new files | Every repeated-content job; a TM without a TMS cannot route the job |
| Machine translation (MT) | Model that generates new output from source text | First-draft stage after TM fills; configured per locale inside the TMS |
| CAT tool | Segment editor where TM leverage, term base, and QA warnings surface | Every human edit session; the translator's working surface |
| Translation management system (TMS) | Router for files, users, deadlines, and audit trail around the other three | Two to three languages or multiple concurrent translators, and every certified lane |
Expanded CAT-tool mechanics live on CAT Tools for Translation: What They Do and Which Ones Translators Actually Use, and the TM side on Memory Translation Software: What It Does Inside a Translator's Workflow.
What does 'system of record' mean for a string in flight?
A string that enters the TMS never leaves it until delivery. The CAT editor and the QA utility reach into it through API calls. Segments are checked out, edited, and checked back in rather than emailed as XLIFF attachments.
flowchart LR A[Source file] --> B[TMS intake] B --> C[TM pre-fill] C --> D[MT first draft] D --> E[CAT edit via API] E --> F[LQA gate] F --> G[Signed attestation] G --> H[Delivery]
When does a translation project actually need a TMS?
A TMS becomes necessary once a project exceeds two or three languages or involves multiple concurrent translators. Two locales with one translator survives on a shared XLIFF. Three or more locales pays the TMS back inside a quarter. Any certified lane makes the TMS mandatory for the audit trail. The calculus is straightforward: a TMS without a translation memory reruns every segment from scratch, and a TM without a TMS cannot route the job to the right vendor or reviewer — the two components are co-dependent, not interchangeable.
| Project shape | Pipeline that works | TMS verdict |
|---|---|---|
| 1 locale, 1 translator | Shared XLIFF, email | Not needed |
| 2 locales, 1 translator per locale | Shared XLIFF, spreadsheet tracker | Borderline |
| 3+ locales, 1 translator per locale | TMS with lane-level TM | Pays back inside a quarter |
| Any locale, concurrent reviewers | TMS with reviewer workflow | Mandatory |
| Any certified lane (USCIS, court, registry) | TMS with audit trail | Mandatory |
Language complexity adds pressure on top of locale count. A single target language can fragment into several regional variants — each carrying distinct approved terminology, spelling conventions, and legal phrasing — that must be stored in separate TMs under the same client account. A TMS is the only practical tool for enforcing that separation at scale, because it applies lane-level metadata at the moment a segment match is retrieved, not after the fact.
Software companies and globally distributed teams are among the first organizations to reach the TMS threshold, because their content ships on release cycles that do not tolerate manual file management. When a product launches across dozens of markets simultaneously, the TMS is what synchronizes vendor queues, MT engine selection, and TM leverage into a single auditable workflow rather than a chain of emails and spreadsheets.
What breaks first when a spreadsheet pipeline scales past the threshold?
Four things break in sequence, and each one compounds the next:
- JSON or XLIFF files edited by hand accumulate version conflicts that no diff tool resolves cleanly across three or more simultaneous locales.
- Translation memory leverage collapses because no central TM reads back — every previously approved segment is invisible to the next translator, so savings that should compound instead vanish.
- The effective per-word rate rises as repeated segments get re-translated; without TM replay, identical strings in a UI or legal document cost full rate on every project cycle.
- Reviewer overwrites become unrecoverable without a lane-level audit trail, which is the precise gap a TMS fills — it logs every segment touch so the named-translator attestation on a certified document has an unbroken chain of custody behind it.
Which lanes make a TMS non-negotiable rather than merely useful?
Two lanes make a TMS non-negotiable, and both share the same underlying reason — they require proof, not just output:
A TMS becomes necessary once a project exceeds two or three languages or involves multiple concurrent translators. Two locales with one translator survives on a shared XLIFF. Three or more locales pays the TMS back inside a quarter. Any certified lane makes the TMS mandatory for the audit trail.
| Project shape | Pipeline that works | TMS verdict |
|---|---|---|
| 1 locale, 1 translator | Shared XLIFF, email | Not needed |
| 2 locales, 1 translator per locale | Shared XLIFF, spreadsheet tracker | Borderline |
| 3+ locales, 1 translator per locale | TMS with lane-level TM | Pays back inside a quarter |
| Any locale, concurrent reviewers | TMS with reviewer workflow | Mandatory |
| Any certified lane (USCIS, court, registry) | TMS with audit trail | Mandatory |
What breaks first when a spreadsheet pipeline scales past the threshold
Four things break in sequence:
5–
What are the core functions of a TMS inside a certified production stack?
A TMS runs 6 core functions inside the production stack: workflow routing, vendor and reviewer orchestration, engine selection, TM and term base storage, connector sync to repositories and CMSes, and program-level reporting. Each function has a named owner on the agency side — the localization project manager, the vendor manager, the LQA lead — not a shared platform login. The pipeline itself is covered Localization Workflow: How Translators and Agencies Wire the Production Pipeline.
| Function | What it does | Named owner |
|---|---|---|
| Workflow routing | Moves files through pre-translation, edit, LQA, delivery | Localization project manager |
| Vendor orchestration | Assigns named translators and reviewers with credentials | Localization vendor manager |
| Engine selection | Picks MT engine per locale and content type | Localization project manager |
| TM and term base storage | Holds segment pairs and glossary per client per locale | Terminology lead |
| Connector sync | Repo and CMS integrations, release-trailing pushes | Localization engineer |
| Reporting | Volume, cost, leverage, cycle time per program | Account director |
How does the TMS orchestrate the TM-then-MT-then-human sequence?
The TMS runs the standard pre-translation sequence automatically. TM fills exact and fuzzy matches first. The configured MT engine handles the remainder. The human post-edits in the CAT editor. TM leverage determines how much of the file each stage touches and how the per-word rate is calculated.
flowchart TD
A[Source segment] --> B{TM match?}
B -- 100% / fuzzy --> C[TM fills segment]
B -- No match --> D[MT generates draft]
C --> E[Human post-edit in CAT]
D --> E
E --> F[LQA sampling]
F --> G[Delivery]
How does the TMS enforce the one-TM-per-client-per-locale governance rule?
The TMS enforces one TM per client per locale with lane-level metadata. The metadata prevents a certified-document segment from overwriting its marketing variant. This is the single most important structural rule for agencies running mixed content types. It is impossible to maintain in a spreadsheet pipeline.
Lane-level metadata tags each segment with:
- Content type — certified document, marketing, product UI, legal, support.
- Source authority — who approved the segment pair.
- Locale variant — en-US vs en-GB kept separate at the TM level.
- Last LQA pass date — for stale-segment flagging.
- Confidentiality flag — segments tied to signed attestations never leak into open pools.
The vendor roster that attaches credentials to each certified-lane signer sits inside the same TMS record. More on that role What a Localization Vendor Manager Does and How to Work With Them.
Where does the TMS hand off to the LQA gate?
The TMS triggers the LQA gate but does not perform it. The gate samples translated output at $40–$90 per 1,000 words against an MQM-style scorecard aligned to ISO 17100. The scorecard covers five dimensions:
- Accuracy — mistranslation, omission, addition.
- Fluency — grammar, spelling, punctuation in the target locale.
- Terminology — glossary adherence against the client term base.
- Locale convention — dates, numbers, currency, address format.
- Style — register match against the client style guide.
Mandatory on every certified lane. A failed sample routes back to the CAT editor through the TMS; a passed sample releases the file for delivery.
What does a TMS NOT do that certified-lane delivery still requires?
A TMS alone does not certify a page, run an LQA scorecard, or produce a named-translator attestation. It is one layer inside a localization solution. The CAT tool, MT engine, TM, QA/LQA gates, connectors, and reviewer workflow are what turn routed strings into deliverable, certifiable output. The full assembly lives on Localization Solution: What Translators and Agencies Assemble to Run Multi-Lane Work.
Why does the TMS not produce the named-translator attestation?
A named-translator attestation is a signed statement by the human translator asserting accuracy and competence. It is the artifact a receiving authority like USCIS accepts. The TMS logs who touched which segment and when. It does not hold the credential, the signature, or the liability.
A specimen certificate of accuracy — the artifact the TMS cannot produce on its own:
I, [Translator name], certify that I am competent to translate from [Source language] into English, and that the attached translation of [Document type] is a true and accurate rendering of the original document to the best of my knowledge and ability.
Signed: [Translator signature]
Date:
Credentials: [Membership / certification number]
Contact: [Translator address or agency letterhead]
- Translator name — the human liable for the translation, not the agency and not the TMS.
- Competence statement — the clause that satisfies USCIS 8 CFR 103.2(b)(3).
- Document type — what is being certified; a birth certificate attestation is not interchangeable with a transcript attestation.
- Signature — wet or e-signature on the human translator, never on the platform.
- Date and credentials — pulls from the vendor record inside the TMS; the TMS routes the data, the human signs the page.
What are the 6 components that sit alongside the TMS in a SaaS localization program?
A SaaS localization program has 6 components alongside the TMS, each run by a named role on the agency side:
- Translation management system — routing and audit trail.
- CAT editor — segment edit surface.
- QA tooling — automated checks for tags, numbers, terminology.
- AI first-draft engine — MT or LLM post-edit entry point.
- Glossary and translation memory — terminology and segment pairs.
- Release-trailing workflow — wired to the client's CI/CD pipeline.
How do agencies choose a TMS without over-buying?
Agencies choose a TMS by matching 4 production constraints to platform features: locale count and concurrency, certified-lane requirements, CI/CD connector needs, and reviewer workflow. Over-buying happens when enterprise features are purchased for 2-locale pipelines that still fit a shared XLIFF. The PM responsibilities that drive configuration sit on How Localization Project Management Works End to End.
What does a translator or small agency evaluate first in a TMS?
Evaluate 3 things first:
- Whether the TMS exposes TM governance at the lane level — one TM per client per locale with metadata.
- Whether it logs an audit trail the certified lane can reference — segment-level who/when records.
- Whether its API lets the chosen CAT tool and LQA utility read and write without duplicate seats.
Which TMS platforms do agencies standardize on?
Agencies standardize on 4 TMS platforms that each cover the production-side stack differently:
- Phrase (formerly Memsource) — cloud-native, strong connector library.
- memoQ Server — translator-friendly, strong on hybrid cloud/on-prem.
- SDL Trados GroupShare — deep CAT integration for Trados shops.
- Lokalise — strong on software strings and developer workflow.
Open-source options like Weblate exist but require in-house ops capacity most freelancers do not carry.
How does a TMS fit TMS-for-LMS, SaaS, and other continuous-localization lanes?
A TMS fits continuous-localization lanes by subscribing to a code repository or CMS, extracting strings on commit, routing them through TM-then-MT-then-human, and pushing translated keys back to the branch. LMS and SaaS clients add a release-trailing workflow wired to the client's CI/CD pipeline. The full continuous model is on Localization SaaS: How Translators and Agencies Run Continuous Localization for Software Clients.
For LMS clients the pattern is narrower: a connector to the LMS content model, locale-specific routing, and an LQA gate before learner-facing publish. The gate matters more here because a mistranslated assessment question propagates to every learner who sees it.
What does the release-trailing workflow look like in practice?
The release-trailing workflow:
- Pulls new or changed keys from the main branch on every commit.
- Assigns them to a locale-specific translator inside the TMS.
- Holds the branch's locale files until LQA passes.
- Merges the translated keys back behind a feature flag.
- Logs the segment-level audit trail against the release tag.
How we run the TMS layer on certified lanes
We treat the TMS as a router and an audit trail, not a translator. Every certified job routes through it so the segment-level log exists when a receiving authority asks for it, and the human translator signs the attestation on the delivered page. The platform does not sign anything on our behalf.
We maintain one TM per client per locale with lane-level metadata that keeps certified-document segments out of marketing pools. Mixing the two once is an incident; mixing them twice is a liability we will not accept.
We refuse to ship raw MT output on a certified lane, even where a client asks us to compress cost. MT is the first-draft stage that feeds the human post-edit; it is not the deliverable. A TMS that routed raw MT past the LQA gate would fail the ISO 17100-aligned scorecard, and the named translator would carry the liability for a page they did not actually translate.
We also refuse to attach a named-translator attestation to any file produced wholly inside the platform. The signature belongs to the human who did the post-edit, and the credentials in the vendor record must match the person on the certificate.
Four questions come up repeatedly from translators and agency ops leads evaluating the TMS layer. They cover: whether a TMS replaces a CAT tool; whether MT makes a TMS redundant; what a TMS costs on the production side; and whether a TMS issues the certification artifact itself.
Frequently asked questions
what is translation management system
A translation management system (TMS) is software that centralizes and automates the workflows involved in producing translations — from job intake and assignment through delivery and archiving. For professional translators and agencies working on certified translations, a TMS handles document ingestion, translator assignment, quality review steps, and final output formatting. Modern SaaS-based platforms also integrate CAT (computer-assisted translation) tools, glossaries, and translation memories so teams can maintain consistency across projects. Some systems now generate first drafts automatically, letting human translators focus on editing, certification, and quality assurance rather than typing from scratch.
what is translation project management
Translation project management is the process of coordinating all tasks, people, and resources required to deliver a translated document on time and to specification. For certified translation workflows this includes receiving source documents, verifying requirements (target language, certification format, deadline), assigning qualified translators, tracking progress, managing review cycles, and issuing the final certified output. Project managers in agencies oversee multiple jobs simultaneously — sometimes dozens per week — so a TMS is essential for avoiding bottlenecks, tracking deadlines, and maintaining audit trails required for certified work.
how does translation software work
Translation software ingests a source document, segments it into sentences or phrases, and presents each segment to a translator alongside suggestions from translation memory, machine translation, or both. In a certified-translation context, the software generates a first draft that a qualified human translator then edits, verifies for accuracy, and certifies. Quality assurance checks flag inconsistencies, untranslated segments, or terminology mismatches against approved glossaries. The final output is formatted for delivery, and the project record — including who translated and reviewed each segment — is stored for compliance and audit purposes.
Related
- ATA Membership: Tiers, Dues, and What Translators Actually Get
- Back Translation: How It Works, When to Use It, and Its Limits
- American Translators Association Certification: Exam, Eligibility, and Credential Recognition
- Sworn Translator: What the Role Is and How Agencies Produce Sworn Translations
- Desktop Publishing in Adobe InDesign: The Translator's Workflow
- Copy Certification by Document Custodian: What Translators and Agencies Need to Know