What is a Translation Memory and how does it work?

A Translation Memory (TM) is a database that stores previously translated text segments (source text and translation) and automatically suggests them for new translation projects. When a new sentence exactly matches (100% match) or resembles (fuzzy match) a stored source sentence, the stored translation is displayed as a suggestion. This saves time and promotes consistency — but always requires human review.

When should I not trust a TM match in IFU translation?

You should critically examine TM matches when: the context has changed (same sentence, different meaning in the new context), the translation originates from a different product area (e.g., orthopedic vs. cardiovascular), regulatory requirements have changed since the original translation, the match is a fuzzy match below 95%, or terminology in the source text has been updated without the TM being cleaned up accordingly.

What should the leverage rate be for medical device IFU translations?

The leverage rate (proportion of text segments with TM matches) varies considerably depending on project type. For initial IFU translations, it typically ranges from 0-20%. For updates to existing IFUs, it can reach 60-80%. For product families with similar IFUs, 40-60% is realistic. Important: while a high leverage rate reduces costs and processing time, it does not replace a full review of every segment.

Technologie

Using Translation Memory Effectively in Medical Devices

Leverage Rates, Fuzzy Matches, and the Limits of Automatic Suggestions

8 min read

What Is Translation Memory?

Translation Memory (TM) is one of the fundamental technologies in the modern translation process. A TM database stores pairs of source and target text segments created during previous translation projects. When a similar or identical source text is detected in a new project, the system automatically suggests the stored translation.

For the medical device industry, Translation Memory offers considerable advantages: it promotes terminological consistency across document versions and product families, reduces processing times for updates to existing IFUs, and lowers translation costs through the reuse of previously reviewed segments. However, these advantages also carry risks when the technology is applied uncritically.

How TM Matching Works

The core of every TM system is the matching algorithm that compares new source sentences with stored entries. The result is expressed as a match rate in percent.

A 100% match (also called an "exact match" or "full match") occurs when the new source sentence exactly matches a stored entry. A "context match" or "101% match" goes further: not only the sentence but also the surrounding context (previous and following sentence) matches. This provides the highest reliability.

Fuzzy matches refer to matches below 100%. A 95% match means the new sentence is 95% similar to the stored one — individual words, numbers, or phrases differ. The lower the match rate, the more manual adjustment is required. In medical technology, the rule of thumb is: fuzzy matches below 75% typically require so much post-editing that a new translation would be more efficient.

Repetitions and Internal Repeats

An often underestimated advantage of TM systems is the detection of internal repetitions. When the same sentence appears multiple times within a document — such as standard warnings or cleaning instructions — it is translated only once and automatically inserted at all locations. This not only ensures consistency but saves considerable time.

Leverage Rates in Medical Technology

The leverage rate describes the proportion of a new project covered by existing TM entries. In medical technology, this rate varies considerably depending on project type.

For initial translation of a completely new IFU, the leverage rate is naturally low — typically 0-20%, unless similar products have already been translated. The greatest savings are achieved with IFU updates: if only 15% of the text has changed, up to 85% can be drawn from the Translation Memory.

Product families also offer high TM potential. If you manufacture three variants of a diagnostic device that differ only in performance data and specifications, large text sections — such as those covering cleaning, maintenance, or troubleshooting — can be reused across products.

Leverage Rate vs. Quality

A high leverage rate is commercially attractive but must not come at the expense of quality. Every TM segment — even a 100% match — must be reviewed. The context may have changed, regulatory requirements may have been updated, or terminology may have been revised. Automatic adoption without review is not acceptable in medical technology.

When You Should Not Trust TM Suggestions

Translation Memory is a powerful tool but not an infallible one. There are situations where TM suggestions can be misleading or even dangerous.

Context-Dependent Translations

The same sentence may need to be translated differently in different contexts. The English sentence "Remove the cap" could refer to a protective cap, a closure, or a cover — depending on the product and context, a different translation is correct. If your TM contains the sentence from a previous project, the suggestion may be wrong for the current context.

Outdated Entries

TM databases grow over years and may contain entries that are no longer correct by current standards. Regulatory changes — such as the transition from MDD to MDR — require systematic revision of the TM. Terms that were correct under the old Directive may be wrong or outdated under the new Regulation.

Cross-Product Risks

When a TM is used across products, there is a risk that product-specific translations are incorrectly transferred to other products. A dosing instruction for an insulin pen system is not transferable to an infusion system, even if the sentences sound similar. In safety-critical areas, this confusion can have fatal consequences.

TM Maintenance and Quality Assurance

A Translation Memory is only as good as its maintenance. Regular upkeep includes cleaning out outdated entries, consolidating duplicate entries with different translations, updating terminology changes, and removing entries originating from erroneous projects.

manualworks integrates TM management directly into the translation workflow. The platform allows TM entries to be tagged with metadata — such as product category, validity period, or quality level — so that translators and reviewers can immediately assess how trustworthy a TM suggestion is.

TM and Machine Translation

Translation Memory and Machine Translation (MT) are different technologies that are increasingly complementary. While TM draws on segments previously translated by humans, MT generates new translations algorithmically. In medical technology, MT is increasingly used as pre-translation, which is then reviewed and corrected by specialist translators (post-editing).

The combination of TM and MT can further increase efficiency: TM matches take priority, and MT is used only for segments without TM matches. For medical technology, however, MT suggestions require even more careful review than TM suggestions, as they are not based on validated human translations.

Conclusion

Translation Memory is an indispensable tool in medical technology for consistency, efficiency, and cost optimization. Yet its strength — the automatic reuse of previous translations — is simultaneously its weakness when applied uncritically. Use TM as a starting point, not as an end result. Invest in maintaining your TM database, train your translators in the critical evaluation of TM suggestions, and ensure that every segment — regardless of match rate — undergoes human review.