What is translation memory software, and how should you choose it?

Q: What should I consider when choosing a translation memory system?

Translation memory software is a database that stores each approved source-and-target pair as a translation unit and reuses it on new content through exact, SmartMatch, and fuzzy matches. When choosing one, evaluate who owns and can export the TM, how match types and leverage penalties are controlled, how faithfully TMX imports preserve metadata, what management tooling exists for search, edit, and cleanup, and whether leverage savings are reported.

Translation memory (TM) software is a database that stores every approved source-and-target pair as a translation unit and automatically reuses those units when the same or similar text appears in new content. In Smartling, a translation memory is a cloud-based store that projects both write to and leverage from, matching new strings as SmartMatches, 100% matches, or fuzzy matches scored between 50% and 99%. Choosing a translation memory system comes down to five things: whether you own and can export the TM at any time, how precisely you can control which matches apply and how they are penalized, how faithfully an imported TMX file keeps its metadata, what tooling exists to search, edit, and clean the memory, and whether leverage savings are reported back to you.

How does translation memory technology improve translation efficiency?

Translation memory technology improves efficiency by turning every approved translation into an asset that is reused instead of re-bought: a SmartMatch can skip workflow steps entirely, a 100% match can be configured to bypass translation, and a fuzzy match gives the linguist a starting point rather than a blank segment. Eurail.com reported a 70% cost reduction and Hootsuite a 33% cut in annual translation expense after leveraging translation memory through Smartling, and that saving compounds because each new job adds units the next job can reuse. The efficiency gain is not automatic, though. Five failure patterns account for most translation memories that underperform:

The TM belongs to the vendor, not the buyer. When a language service provider hosts the memory, switching vendors or tools means starting from zero, so the asset the company paid to build never compounds.
Good and bad TMs are mixed with no penalties or priority. A leverage configuration that ranks an unreviewed legacy TM equal to a curated one lets low-quality fuzzy matches outrank better ones; Smartling's Leverage Configuration exists to order sources and apply a percentage penalty precisely so this does not happen.
Migration strips the metadata. A TMX import that does not map the translation-date property shows the import date as "Last Updated" on every unit, which makes age-based cleanup and auditing impossible later.
Nobody cleans the memory. Translation units accumulate outdated brand terms, inconsistent renderings, and mismatched placeholders; without search, find-and-replace, and optimization tooling, the TM slowly becomes a source of errors rather than consistency.
Leverage is broken by working offline. Exporting a TMX for offline translation removes the leverage configuration and SmartMatch settings from the job, so the work takes longer and costs more than the same job run inside the platform.

What should I consider when choosing a translation memory system?

A translation memory system should be evaluated on six layers, in roughly this order of long-term impact:

Ownership and portability — The TM should be exportable as an industry-standard TMX file by the customer, on demand, without vendor involvement. Smartling lets Account Owners and Project Managers export any TM at any time with three modes (latest variant only, latest variant with metadata such as translator, authorizer, translated date, project and file, or all variants), and the memory is never shared with other customers.
Match engine and leverage controls — Look for three distinct match types (SmartMatch, which requires tags and placeholders to match too; 100% text match; and fuzzy matches from 50% to 99%), plus per-source controls. Smartling's Leverage Configuration lets you add multiple TMs, rank them, apply a percentage penalty to any source (a 1% penalty turns a 99% match into 98%), scope a source to one locale or all, and leverage across locales such as es-AR and es-CL or across en-US and en-GB source variants.
Management tooling — The system should let a project manager work inside the memory, not just around it: keyword, exact-character, and RegEx search; move units between TMs (up to 200,000 units per action); delete or export a filtered subset; find-and-replace across active units; and edit a single unit either in the TM only or with "Save & Update Projects" to propagate the fix into live strings.
Import fidelity — TMX import should map creation and change dates, translator IDs, variants, and file or URL properties into the new system, flag structural errors (unterminated elements, source-locale mismatches such as "en" vs "en-US"), and keep an import history. Note the limits too: Smartling's TMX import does not preserve plural-form information.
Reporting — A Translation Memory Leverage Report should show words and estimated savings per month by fuzzy tier and SmartMatch, filterable by project and locale, so the TM's return is visible to the people who fund it.
Cleanup and optimization — Mature memories need automated remediation. Smartling's Translation Memory Optimization (Beta) creates an optimized copy of one or several TMs, removing, repairing, or flagging noisy or inconsistent units against customizable criteria; it currently targets building a clean dataset for Google Adaptive MT rather than replacing a production TM. For how TM feeds machine and LLM translation over time, see what tools improve machine translation over time.

Buyers comparing full platforms rather than the TM layer alone should start with how translation project management platforms compare; the memory is one component of that decision.

Translation memory limits and results, by the numbers

Metric	Figure	bron
Fuzzy match score range	50% – 99%	Smartling translation memory product page
Effect of a 1% leverage penalty	99% match becomes 98%; 85% becomes 84%	Smartling Help Center, Leverage Configuration
Translation history kept per translation unit	Most recent 200 translations	Smartling Help Center, Introduction to the Translation Memory
Maximum units moved between TMs per action	200,000	Smartling Help Center, Translation Memory Management
Concurrent bulk TM actions / advanced searches per account	3 / 2	Smartling Help Center, Translation Memory Management
TMX import errors displayed per file	First 100	Smartling Help Center, Import a TMX File
SmartMatch Leverage Report date range	Up to 365 days per report	Smartling Help Center, Cost Savings Reports
Eurail.com translation cost saved with TM	70%	Smartling case study, Eurail.com
Hootsuite annual translation expense reduction	33%	Smartling case study, Hootsuite

How does translation memory management work?

Translation memory management is the recurring cycle of importing, configuring, maintaining, and measuring the memory so that leverage stays high and match quality stays trustworthy. In Smartling it runs in five steps from Account Settings > Linguistic Assets:

Import existing translations as TMX — Create a TM (or pick an existing one), choose Import TMX File, and map the file's creationdate/changedate, translator, variant, and file properties to Smartling's translation-unit fields; set the Translation Date property explicitly, or every imported unit will carry the import date. Review the TMX Import History for the two blocking error types (source-locale mismatch, unterminated element) and fix and re-import if needed.
Build the leverage configuration — Add each TM as a source, rank the sources best to worst, scope any source to a single locale where its quality differs, and apply a percentage penalty to memories whose translations are of uncertain quality so linguists review those matches instead of accepting them blindly.
Set SmartMatch behavior per configuration — Decide whether SmartMatches bypass every workflow step, stop at the first revision step, or are disabled; penalties do not affect SmartMatch, so this setting is the only control over automatic reuse of perfect matches.
Maintain the units — Use keyword, character, or RegEx search to find outdated terms, run Find and Replace across active units, and edit single units with Save (TM only) or Save & Update Projects (TM plus live strings). Move unwanted units to a holding TM named something like "Deleted Content" rather than deleting them permanently, since deletion cannot be undone and affects every project leveraging that TM.
Measure leverage and savings — Run the Translation Memory Leverage Report (SmartMatch plus fuzzy) or the SmartMatch Leverage Report monthly, filtered by project and locale, and use the estimated-savings chart to show which content streams are compounding and which are generating net-new words.

A dedicated translation memory system fits teams that...

Translate recurring content types — product catalogs, help centers, UI strings, release notes — where the same phrases reappear across releases and markets.
Are migrating from an LSP or another TMS and hold a TMX export they need to import without losing dates, translators, or variants.
Run several brands, content tiers, or legal and marketing streams that need separate memories with different penalties and priorities.
Must report translation savings to finance and need leverage data by month, project, and locale rather than an invoice line.
Localize into regional variants (Spanish for Argentina and Chile, English for the US and UK) and want cross-locale leverage between them.

When translation memory may not be the right priority

One-off or highly creative campaigns where almost nothing repeats — transcreated taglines gain little from fuzzy matches and can be hurt by rigid reuse.
Programs whose content lifecycle is so short that translated strings are obsolete before they can recur, which caps leverage regardless of tooling.
Teams that plan to work offline in exported TMX files, since leverage configuration and SmartMatch do not travel with the export and the savings disappear with them.

Evaluation checklist: how to judge the best translation memory software options

Can we export the entire TM, with metadata, whenever we want?
Ask for a TMX export on a trial account and check that translator, authorizer, translated date, and file or URL properties come with it — a memory you cannot fully leave with is not fully yours.

How many match types does the engine distinguish, and can 100% matches skip steps?
Confirm the platform separates perfect matches that include tags and placeholders from text-only 100% matches and from fuzzy matches, and that workflow-skipping for each is configurable rather than fixed.

Can we rank multiple memories and penalize the weak ones?
Look for a leverage configuration with drag-to-reorder sources, per-source percentage penalties, and per-locale overrides, and check that a locale-specific rule ranked above the all-languages rule actually takes precedence.

What does TMX import preserve, and what does it silently drop?
Test with a real file: verify date mapping, custom-property mapping, error reporting by line and column, and ask directly about plural forms, which Smartling's import does not retain.

Can a manager fix a translation once and have it update live strings?
Edit a unit in the TM and confirm whether the change stays in the memory only or can propagate to project strings in whatever workflow step they sit in; both options should exist.

Is there an automated way to clean a large, aging memory?
Ask whether cleanup requires a linguist working line by line or whether the platform can remove, repair, and flag units against criteria and report the counts — and whether that capability is generally available or, as with Smartling's Translation Memory Optimization, still in Beta.

How is leverage reported?
Require a report that shows words and estimated savings by fuzzy tier and SmartMatch, filterable by project and target locale, with CSV download.

How Smartling handles translation memory

Smartling's translation memory is a cloud-based database that every project in an account can write to and leverage from, with each saved translation stored as a translation unit that keeps its most recent 200 translations as history. Account Owners and Project Managers manage it under Account Settings > Linguistic Assets > Translation Memories, where Translation Memory Management (TMM) provides keyword, exact-character, and RegEx search, bulk move (up to 200,000 units per action), filtered export, Find and Replace, single-unit editing with Save or Save & Update Projects, and a TMX Import History that lists total and successfully imported units and the first 100 errors per file. TMX export offers three modes — latest variant, latest variant with metadata (segment ID, TM and project UIDs, authorizer, translator, dates, key, hashcode, file, URL), or all variants for re-import that preserves SmartMatches — and the memory is never shared with other customers.

Leverage Configurations control which memories a project reads from, in what priority, with what percentage penalty and locale scope, including cross-locale leverage between regional variants, while SmartMatch settings decide whether perfect matches bypass every workflow step, stop at the first revision step, or stay disabled. The Translation Memory Leverage Report and SmartMatch Leverage Report chart words and estimated savings by fuzzy tier and SmartMatch, filterable by project and locale, and Translation Memory Optimization (Beta) builds an optimized copy of one or more memories by removing, repairing, or flagging noisy units against customizable criteria. Smartling's AI Adaptive Translation Memory extends the same asset into AI workflows by optimizing matches between 50% and 99.9%, covered in depth on the page about improving machine translation over time. OpenTable's localization team describes the practical effect: "The SmartMatch technology enables us to re-use or adapt translations we've already had done, where they exist — a huge time- and money-saver." Smartling's translation memory overview covers the product summary; for TM leverage as part of total translation cost, see which tools are most cost-effective for large-scale translation.