Translation Memory and Glossaries: The Hidden Infrastructure Behind Consistent Multilingual Content
Every time you translate the same phrase differently, you confuse your audience and waste money. Translation memory and glossaries are the unsexy infrastructure that makes multilingual content consistent, efficient, and scalable.
Consistency is the most undervalued quality in translation
Consistency is the most undervalued quality in translation. A viewer who watches three of your videos in Spanish and encounters three different translations for the same product feature does not think you have a diverse vocabulary — they think you are disorganised or, worse, talking about three different things. For brands, consistency in translated content is a trust signal. For educational content, it is a comprehension requirement. For technical content, it is a safety necessity.The tools that make consistency achievable at scale are translation memory (TM) and terminology glossaries. These are not glamorous technologies. They do not generate headlines or demo well at conferences. But they are the foundation that separates professional multilingual operations from chaotic ones.What Is Translation Memory?Translation memory is a database that stores previously translated segments — typically sentences or sentence-like units — alongside their source text. When a new text contains a segment that matches or closely matches something already in the TM, the stored translation is suggested to the translator.
If the match is exact, the translator can accept it without modification
If the match is exact, the translator can accept it without modification. If it is close (a fuzzy match), the translator uses it as a starting point and edits as needed.How It Works in PracticeImagine you translated a product description six months ago that included the sentence: 'Zingo uses neural machine translation to deliver real-time video subtitles.' This sentence and its translation are stored in TM. When you translate a new blog post that contains the same sentence, the TM recognises it and auto-populates the translation. If the new text says 'Zingo uses neural machine translation to deliver real-time audio subtitles' (one word different), the TM flags it as a fuzzy match and presents the stored translation with the difference highlighted. The translator updates the single word rather than translating the entire sentence from scratch.BenefitsConsistency: The same source text produces the same translation every timeSpeed: Translators do not retranslate content they have already translatedCost reduction: Exact matches are typically billed at a lower rate than new translation, and fuzzy matches at an intermediate rateQuality: Reusing approved translations reduces the opportunity for errorsLimitationsTM works at the segment level, which means it does not capture document-level context. A sentence that was perfectly translated in one context might need a different translation in another context, but TM will suggest the stored version regardless. Human judgment is still needed to assess whether a TM match is appropriate in the current context.TM also needs maintenance. A translation that was correct three years ago may use outdated terminology or reflect an old brand voice. TMs should be periodically reviewed and cleaned to ensure that stored translations reflect current standards.What Are Terminology Glossaries?A terminology glossary (or termbase) is a structured list of terms and their approved translations in each target language.
Implementation Guide
Unlike TM, which works at the sentence level, glossaries work at the term level — individual words or short phrases that must always be translated the same way.What Belongs in a GlossaryProduct and feature names: How you refer to your product and its features in each languageBrand-specific terms: Proprietary terminology, slogans, and branded conceptsTechnical terms: Industry-specific vocabulary with approved translationsDo-not-translate terms: Terms that should remain in the source languagePreferred translations: Where multiple valid translations exist, the preferred option for your contextGlossary StructureEach glossary entry should include:Source termTarget term for each languageDefinition or context (so translators understand the term's meaning)Usage notes (any restrictions on when this translation applies)Domain (the subject area where this term applies)Status (approved, provisional, deprecated)Why Glossaries Matter More Than You ThinkA glossary is your brand's linguistic identity in each language. Without one, every translator makes independent decisions about how to translate your terminology, and those decisions will vary. One translator renders your feature as 'instant dubbing,' another as 'real-time dubbing,' a third as 'immediate dubbing.' All are correct translations, but the inconsistency undermines the brand.For video content specifically, inconsistent terminology is even more noticeable than in text. When viewers hear a term one way in one video and a different way in the next, the inconsistency registers immediately in audio. In text, a reader might not notice the variation. In audio, they almost certainly will.Building Your Translation AssetsStarting a Translation MemoryIf you have never used TM, start by collecting all existing translations. Previous subtitle files, dubbed scripts, translated marketing materials, localised website content — all of this can be aligned with the source text and imported into a TM. The initial investment in aligning past translations pays off immediately as the TM starts suggesting matches for new content.If you are starting from zero with no existing translations, your TM builds organically as you translate new content. Each translated project adds to the TM, and the benefits accumulate over time.
The first project gets no TM leverage
The first project gets no TM leverage. The tenth project might see 30-40% of its content matched by TM.Building a Glossary from ScratchStart with your most visible terminology — product names, feature names, and any terms that appear frequently in your content. For each term, determine the approved translation in each target language. This initial glossary can be modest — 50-100 terms is a solid starting point for most organisations.Involve native speakers from each target market in the glossary creation process. A term that seems obviously correct to a bilingual person in your office may not be the most natural or commonly understood term in the target market. Test proposed terms with actual users when possible.Maintaining Translation AssetsTranslation assets are living documents that need regular maintenance:Add new terms as your product evolves and new concepts emergeReview and update quarterly for accuracy and relevanceRetire deprecated terms that no longer applyClean TM by removing outdated translations or correcting known errorsAudit consistency by checking that all recent translations follow the glossaryTools and TechnologyComputer-Assisted Translation (CAT) ToolsCAT tools are the software platforms that house TM and glossaries and integrate them into the translation workflow. The major CAT tools — SDL Trados, memoQ, Memsource (now Phrase), Smartling — all support TM and glossary management alongside the translation interface. When a translator works in a CAT tool, TM matches and glossary terms are displayed automatically, guiding consistent translation without manual lookup.Cloud-Based TMModern TM systems are cloud-based, allowing multiple translators working in different languages and locations to draw from and contribute to the same TM simultaneously. This is essential for teams localising content into multiple languages concurrently — a translation completed in Spanish can inform the German and French translations happening at the same time.AI and TM IntegrationAI translation engines are increasingly integrated with TM systems.
Tips for Success
The workflow becomes: check TM for exact and fuzzy matches first, use AI translation for segments with no TM match, then have the human translator review and approve. This hybrid approach maximises efficiency while maintaining quality. It also feeds the TM — every human-approved translation is stored for future use.Applying This to Video TranslationFor video content specifically, TM and glossaries serve several unique functions:Script ConsistencyWhen translating a series of videos — a course, a content series, a product video library — TM ensures that explanations and descriptions are consistent across the series. A concept explained in video 3 should use the same terminology in video 7.Subtitle ReuseIf you update a video but keep some segments unchanged, TM allows you to reuse the existing subtitle translations for unchanged segments and only translate the new content. This saves time and ensures visual consistency between the old and new versions.Voiceover TerminologyGlossaries ensure that voice actors and AI voice systems use consistent terminology across all content. Provide the glossary to voice actors before recording sessions and to AI dubbing platforms as a terminology constraint.Cross-Format ConsistencyYour video subtitles, website copy, app interface, and marketing materials should all use the same terminology. Sharing TM and glossaries across these formats ensures that a user who reads your website in Spanish and then watches your tutorial video in Spanish encounters consistent language throughout.Measuring the ROICost savings: Track the percentage of TM matches in each project. At 40% match rate with a 50% discount on matched segments, you save 20% on translation costsTime savings: TM matches are faster to process, reducing turnaround time proportionallyError reduction: Track translation errors over time. Organisations with mature TM and glossary assets typically see error rates decrease as the assets improveConsistency scores: Audit translations for terminology consistency and track improvement over timeTranslation memory and glossaries are not one-time investments. They are compounding assets that become more valuable with every project. The organisation that starts building these assets today will have a significant quality and cost advantage over competitors who wait.