OnlineDocTranslatorGuides

Answer

Memory Translation Software: What It Does Inside a Translator's Workflow

Memory translation software stores approved source–target segment pairs and replays them on new jobs, cutting re-translation work and enforcing consistency

Working translators and agencies run their production around a translation-memory layer buried inside a CAT tool — a database that remembers every approved segment and replays it on the next job. That layer is what the industry calls memory translation software, and it governs cost, consistency, and audit trail across every release.

Key points

  • Memory translation software is not a standalone product — it is the translation-memory layer inside a CAT tool such as Trados, memoQ, Phrase, or OmegaT, functioning as a database that stores approved source-and-target segment pairs and replays matches automatically on new content.
  • The threshold model translators negotiate with clients recognises three tiers: a fuzzy match floor at 75% similarity, a 100% exact match where the new segment is identical to a previously translated one, and a 101% in-context match where both the segment and its surrounding segments also match — the highest-value tier.
  • Translation memories are exchanged between tools using TMX (Translation Memory eXchange), an XML-based format that lets linguists and agencies move TM assets across platforms without data loss.
  • Professional translators cluster around seven CAT tools with built-in TM in 2025: SDL Trados Studio, memoQ, Phrase, Smartcat, Smartling, OmegaT, and Wordfast Pro.
  • Pricing spans free (OmegaT, open-source) to
    • Memory translation software is not a standalone product — it is the TM layer inside a CAT tool such as Trados, memoQ, Phrase, or OmegaT.
    • The threshold model translators negotiate with clients is 75% fuzzy, 100% exact, and 101% in-context (preceding and following segments also match).
    • Professional translators cluster around 7 CAT tools with built-in TM in 2025: SDL Trados Studio, memoQ, Phrase, Smartcat, Smartling, OmegaT, and Wordfast.
    • SDL Trados Studio runs $400–$900 perpetual, memoQ $620–$900 perpetual, Phrase $27–$55 per user per month, OmegaT free, Wordfast lower than Trados.
    • Governance rule: one TM per client per locale, with lane metadata that stops a certified-document segment from overwriting its marketing variant.
    • The LQA gate downstream runs at $40–$90 per 1,000 words against an MQM-style scorecard aligned to ISO 17100.

    7–$80 per user per month (Phrase, cloud), $400–$900 perpetual (SDL Trados Studio), and $620–$900 perpetual plus annual support (memoQ); Wordfast Pro sits below Trados on price for a cross-platform desktop option.

  • Governance rule: one TM per client per locale, with lane-level metadata that prevents a certified-document segment from overwriting its marketing variant — the single most important structural rule for agencies running mixed content types.
  • The LQA gate downstream runs at $40–$90 per 1,000 words and scores output against an MQM-style scorecard aligned to ISO 17100; it is mandatory on certified lanes.
  • The standard pre-translation sequence is: TM fills exact and fuzzy matches first, a configured MT engine handles the remainder, then a human post-edits — TM leverage determines how much of the file each stage touches.

What is memory translation software?

Memory translation software is the translation-memory layer inside a CAT tool — a database that stores every approved source-and-target segment pair from past projects and automatically replays matches on new source segments, so a translator edits reuse instead of retyping it. Each commit on a confirmed segment writes the pair back to the database. The asset grows with every job and compounds across releases.

Buyers confuse TM with three adjacent things. The table below separates them.

TermWhat it isWhen you need it
Translation memory (TM)Database of approved source-target segment pairsAny repeat content: software strings, legal boilerplate, product releases
Machine translation (MT)Model that generates new target text from sourceSegments the TM cannot fill; post-editing lanes
Term base (TB)Glossary of approved terms with forbidden alternatesBrand names, regulated phrasing, product terminology
TMSTranslation management system that routes files, users, deadlinesOrchestrating multi-vendor, multi-locale projects

Is it a standalone product or a feature inside a CAT tool?

Memory translation software is a feature layer, not a standalone product. Translators access it through CAT Tool: What It Is, What It Does Inside a Translator's Workflow, where TM runs alongside the term base and QA checker on every segment. The CAT tool reads bilingual files — SDLXLIFF, MQXLIFF, XLIFF 1.2 or 2.0 — and exchanges TM data through TMX for interchange between tools.

How is it different from machine translation and from a TMS?

TM replays human-approved segment pairs. Machine translation generates new output from a model. A TMS routes files, users, and deadlines around both. Confusing the three is the single most common buying mistake translators report. A TMS without a TM reruns every segment. A TM without a TMS cannot route the job.

How does memory translation software work on a live project?

On every new segment, the CAT tool queries the TM database and pre-fills the target cell when a match crosses the agreed threshold. The typical floor is 75% for fuzzy matches, 100% for exact matches, and 101% for in-context matches where the preceding and following segments also match. These thresholds are not just software defaults — they are the lane-level operating contract translators negotiate with clients, written into statements of work.

flowchart TD
 A[New source segment] --> B{Query TM}
 B -- 101% in-context --> C[Pre-fill, treat as approved]
 B -- 100% exact --> D[Pre-fill, pass through QA]
 B -- 75-99% fuzzy --> E[Suggest, flag for edit]
 B -- Below 75% --> F[Send to MT or human]
 C --> G[Commit, write back to TM]
 D --> G
 E --> G
 F --> G

What happens at 75%, 100%, and 101% matches?

At 75% the translator sees a fuzzy suggestion flagged for edit; at 100% the exact prior translation pre-fills but still passes through QA; at 101% the in-context match pre-fills and is treated as approved on most lanes. Match-rate discounts tie directly to this ladder on client POs:

  • 75–84%: light edit expected, 60–70% of full rate
  • 85–94%: medium edit, 50–60% of full rate
  • 95–99%: light proofread, 30–40% of full rate
  • 100% exact: QA review only, 20–30% of full rate
  • 101% in-context: approved without edit, 0–15% of full rate

How does pre-translation combine TM with machine translation?

Pre-translation runs as a two-engine sequence: TM fires first and fills every exact and fuzzy match, then the configured MT engine drafts the remainder. The order matters. It pushes per-word spend toward review instead of first-pass translation. TM-then-MT is the production default inside modern CAT tools, not a theoretical option.

What does a translator actually do inside the segment grid?

The translator confirms or rewrites each pre-filled segment, resolves term-base flags, clears QA warnings on tags and numbers, and commits. Every commit writes back to TM. The core production steps:

  1. Open the bilingual file in the CAT tool's segment grid.
  2. Review the pre-filled target from TM or MT.
  3. Edit against the term base and style guide.
  4. Clear tag, number, and spacing QA warnings.
  5. Confirm the segment — the pair writes back to TM.
  6. Export the finalised bilingual file to the delivery format.

Why us

How much work does TM leverage actually remove?

Leverage depends heavily on content type. Software strings, menu items, error codes, and legal boilerplate reuse heavily between releases because the underlying content changes minimally from version to version. Marketing copy reuses lightly because it is deliberately rewritten each campaign. On mature software products, TM fills the majority of segments before a translator opens the file. Translation quality and brand consistency both improve when translating business assets with repetitive content, precisely because the memory layer locks in the wording that already passed review. Content types ranked by typical TM leverage:

  • UI strings and error codes: 70–95% leverage on point releases
  • Technical documentation: 40–70% leverage per revision
  • Legal and regulatory boilerplate: 60–90% leverage
  • Product descriptions: 20–40% leverage
  • Marketing campaigns: 5–15% leverage

How does TM protect brand voice and regulated phrasing?

Because the TM stores the approved target segment verbatim, every reuse ships the exact wording that passed review the first time. Brand voice, product names, and regulated phrasing stay stable across every subsequent locale and release without additional review cycles. A drug-safety disclaimer approved once keeps its exact wording on release 14. Consistency is not incidental — it is structurally guaranteed by the match mechanism itself, making TM-backed output inherently more defensible regulated industries than fresh translation on each project.

Which memory translation software do professional translators actually use in 2025?

Professional translators cluster around 7 CAT tools with built-in TM in 2025: SDL Trados Studio, memoQ, Phrase, Smartcat, Smartling, OmegaT, and Wordfast. The choice turns on deployment model, bilingual format, and per-seat pricing against the lane each translator bills.

Pricing and deployment snapshot (US, 2025)

The table below restates the site's canonical pricing bands. Deeper comparison lives CAT Tools for Translation: What They Do and Which Ones Translators Actually Use.

ToolDeploymentBilingual formatUS 2025 pricingLane fit
SDL Trados StudioDesktopSDLXLIFF$400–$900 perpetualAgency vendor pools, certified lanes
memoQDesktop + serverMQXLIFF$620–$900 perpetual + annual supportAgency workhorse, LiveDocs reuse
PhraseCloud-nativeXLIFF 1.2 / 2.0$27–$55 per user per monthDistributed teams, software strings
SmartcatCloud-nativeXLIFFFree tier + paid plansBrowser-based freelancers, marketplace
SmartlingCloud-native TMSXLIFFEnterprise contractDistributed enterprise content teams
OmegaTDesktop (cross-platform)XLIFF, TMXFree (open-source)Open-source freelance budgets
Wordfast ProDesktop (Windows/macOS/Linux)TXLF, XLIFFLower than TradosCross-platform freelance throughput

Is there free memory translation software worth running?

OmegaT is the free, open-source option translators actually ship work on. It offers TM, term base, and QA checks but lacks signed-export tooling needed for some certified lanes. The gaps and reach of OmegaT are detailed OmegaT for Professional Translators: What It Does and What It Lacks. Smartcat ships a free tier with cloud TM for individual freelancers.

Which tool fits which lane?

Trados and memoQ dominate agency vendor pools and certified lanes. Phrase and Smartling fit distributed teams and continuous software strings. Smartcat suits browser-based freelance marketplaces. OmegaT and Wordfast cover budget and cross-platform freelance work. Trados-specific trade-offs sit Trados as a CAT Tool: Features, Costs, and Honest Trade-offs; Phrase daily workflow sits Phrase CAT Tool: Features Translators Use Every Day.

Pricing

How accurate is memory translation software for professional use?

Memory translation software is accurate only to the quality of the segments already approved into it. A 100% match replays verbatim. A mistranslation in the TM becomes a mistranslation on every reuse. Accuracy is governed downstream by the LQA gate, not by TM itself.

Where does TM accuracy break down in practice?

TM accuracy breaks down in 4 recurring places:

  • Morphology-poor exact matches that need declension edits in Slavic, Finnish, or Hungarian targets.
  • Locale drift between en-US and en-GB, or between es-ES and es-MX, where a 100% match ships the wrong variant.
  • Tag-dense DOCX and IDML segments that score 100% but still need placeholder repair around formatting runs.
  • Stale segments written for an older product name, deprecated feature, or prior regulatory wording.

How does the LQA gate catch TM errors before publish?

LQA samples translated output at $40–$90 per 1,000 words against an MQM-style scorecard aligned to ISO 17100. The scorecard scores defects across accuracy, fluency, terminology, locale convention, and style. It blocks publish on certified lanes if the sample fails. LQA sits downstream of TM leverage on every certified job.

Computer-Assisted Translation Tools: The Working Translator's Field Guide

How does memory translation software handle context across segments and locales?

Context is handled on 3 axes: the 101% in-context match that only fires when the preceding and following segments also match, concordance search that lets a translator query the TM for a phrase in surrounding text, and lane-level metadata that separates TMs by locale and content type.

Why does one TM live per client, per locale, with lane metadata?

One TM per client per locale carries lane-level metadata so a certified-document segment does not overwrite a marketing variant of the same source string. That separation protects tone, register, and legal accuracy — a failure mode that merged TMs silently create. Lane metadata tags each segment with:

  • Content lane (certified, legal, marketing, UI, support)
  • Source product or document type
  • Approver and LQA pass date
  • Register (formal, neutral, promotional)

What is concordance search and when do translators run it?

Concordance search queries the TM for any string, not just the current segment, returning every past bilingual pair that contains the phrase. Translators run it to resolve ambiguous terms, check prior client-approved phrasing, and reconcile competing fuzzy suggestions. It is the manual counterpart to automatic matching.

What are the limitations of memory translation software today?

Memory translation software has five known limitations that every translator and agency manager should understand before treating TM leverage numbers as unconditional gains:

  1. Zero coverage on new content. TM only reuses what has already been approved — on a first-time project or a heavily rewritten release, leverage is near zero and every word is billed at full rate.
  2. Penalises creative copy. TM rewards repetitive, templated content and offers little value on marketing campaigns or editorial copy that rewrites every release by design; the 5–15% leverage typical of campaigns means TM ROI is minimal in that lane.
  3. Requires disciplined curation. As product names, regulatory wordings, and brand guidelines change over time, stale segments accumulate in the database. A TM that is not actively maintained drifts out of sync with the current approved terminology.
  4. Struggles with heavily tagged formats. DOCX, IDML, and XML files with dense inline tagging can fracture natural sentences into multiple short segments, reducing match rates and introducing tag-error risk when matches are applied.
  5. Inherits past QA defects at scale. A segment approved under weaker QA — or by a translator whose work was later found to be substandard — replays silently on every 100% match. One defective segment approved years ago can propagate across hundreds of pages in subsequent releases.

When does a TM become a liability instead of an asset?

A TM becomes a liability when it contains segments tied to an older brand name, a superseded regulatory phrase, or output that never passed a proper LQA gate. Because 100% matches replay without prompting translator review, they inherit every past defect at full scale and at full speed. The silent nature of exact-match reuse is precisely what makes TM governance non-negotiable on any production-grade deployment: without it, quality problems compound across releases rather than being caught and corrected.

How do agencies curate TMs over time?

Effective TM curation is an ongoing operational discipline, not a one-time setup task. Agencies that maintain clean, high-value TM assets run four recurring actions:

  • Lane separation: keeping one TM per client per locale, with lane-level metadata that prevents certified-document segments from overwriting marketing variants — the structural rule that stops a legal disclaimer from contaminating a tagline database, and vice versa.
  • Versioned snapshots: maintaining dated TM exports so that any problematic batch of segments is rolled back without destroying the broader asset.

  • Segment retirement: actively identifying and purging segments tied to deprecated product names, discontinued SKUs, and superseded regulatory language before those segments are promoted to new projects.
  • Post-import LQA sampling: re-running LQA scoring after any mass TM import, migration, or agency merger to validate that incoming segments meet current quality thresholds before they are eligible for exact-match reuse.

How we run memory translation software on certified lanes

We treat the TM as a production asset with legal weight, not a convenience cache. One TM per client per locale, lane metadata on every segment, and an LQA sample before anything on a certified lane ships. The TM-then-MT sequence runs as the default pre-translation order so review spend lands where it reads.

We refuse to merge a certified-lane TM into a marketing-lane TM to inflate leverage numbers. The downstream failure — a marketing paraphrase overwriting a regulated disclaimer — is silent until the receiving authority catches it, and by then the segment has shipped across 20 jobs. The leverage gain is not worth the audit risk.

We also decline to run 100% matches through without QA on certified lanes, even where the client PO permits a 0% rate. A 100% match on a stale segment is the single most common source of silent regression we see. The cost of a 2-minute QA sweep beats the cost of a rejected filing. For how this sits inside the broader offering, see Service Localization: What Translators and Agencies Actually Deliver.

The 4 questions translators ask most about memory translation software cover the best tool, how it handles context, whether it is accurate enough for paid work, and where it still falls short.

Frequently asked questions

What is translation memory?

Translation memory (TM) is a database that stores previously translated sentence pairs — a source segment alongside its approved target-language equivalent — so translators can reuse them automatically on future projects. When new source text arrives, the TM engine searches for segments that match or closely resemble stored entries, returning exact matches (100%) or fuzzy matches (typically scored 50–99%). This eliminates re-translating identical or near-identical content, cutting both turnaround time and per-word cost, which is especially valuable in certified translation workflows where consistency in legal terminology is non-negotiable.

What is translation memory software?

Translation memory software is a computer-assisted translation (CAT) tool that manages a TM database and presents stored segment matches to a translator inside a side-by-side editing environment. Mainstream platforms — such as SDL Trados Studio, memoQ, Phrase (formerly Memsource), and Wordfast — combine a TM engine with termbases, quality-assurance checks, and project management features. For agencies producing certified translations, TM software enforces consistent rendering of proper names, legal clauses, and boilerplate across documents, reducing the risk of errors that could invalidate a certification.

How does memory translation software handle context across languages?

Modern TM software handles cross-language context primarily at the segment level, matching source sentences against stored pairs and applying glossary constraints to enforce consistent terminology. More advanced tools add context matching — storing not just the segment but the segments immediately before and after it — so a 100% match in a known surrounding context (called an "in-context exact" or ICX match) is treated as more reliable than an isolated 100% match. For certified translation, this matters because a term like "legal guardian" may translate differently depending on whether the surrounding document is a court order or a medical consent form.

Is memory translation software accurate enough for professional use?

Yes — translation memory software is a professional-grade tool used as the production backbone at the vast majority of language service providers worldwide. Its accuracy for certified translation depends on the quality of the TM itself: a well-curated TM built from reviewed, certified documents can supply near-publication-ready segments for high-match content, while low-quality or mixed-domain TMs introduce errors. Industry standard workflow keeps a qualified human translator reviewing every segment, and certified translations additionally require a credentialed professional to sign off on the final output regardless of match scores.

What are the limitations of current memory translation software?

Translation memory software struggles with low-repetition content, highly inflected languages, and long-distance grammatical dependencies that cause stored segments to fit poorly in new contexts. Fuzzy matches below roughly 75% often require so much editing that reuse provides little efficiency gain. TMs also grow stale: outdated entries can propagate incorrect terminology into new certified documents if the database is not actively maintained and versioned. Additionally, TM alone provides no machine-generated first draft for zero-match segments — that gap is filled by integrating MT or, for certified-translation workflows, a dedicated first-draft generation layer.

Related