// DEEP DIVE | AI TRANSLATIONS
Translation Memory: Your Company Asset for Consistent Translations
A Translation Memory System (TMS) stores every translated passage as a segment pair. When the same sentence or a similar one comes up again, the system proposes the stored translation. What sounds banal is the single biggest cost lever in localization: for manuals, T&Cs, and product catalogs, 30 to 60 percent of segments repeat. Without TMS you pay for them every time. With TMS, you pay once.
THE PROBLEM
Why pay for the same translation three times?
Bei technischen Handbüchern, Datenblättern, AGB oder Produktkatalogen ändern sich von Version zu Version oft nur kleine Teile. Eine neue Sicherheitswarnung hier, ein geändertes Material dort. Der Rest des Dokuments ist identisch zum Vorjahr. Trotzdem schickt die meisten Unternehmen das komplette Dokument zur Übersetzung und zahlen es nach Wortpreis komplett neu.
Ohne Translation Memory hat das Übersetzungsbüro keine Möglichkeit zu wissen, dass Satz 47 dieselbe Bedienungsanleitung-Passage ist, die im Vorjahr schon übersetzt wurde. Er wird neu übersetzt, neu abgerechnet, und im schlimmsten Fall noch leicht anders formuliert. Was inkonsistente Terminologie nach sich zieht, die der Kunde später zu spüren bekommt.
Mit einem strukturierten TMS landet jede Übersetzung als Segment-Paar in einer Datenbank, die Ihnen gehört. Beim nächsten Auftrag wird zuerst gegen das TMS geprüft, Match-Raten werden ermittelt, nur die echten Deltas gehen in die Neuübersetzung. Konsistenz bleibt erhalten, die Rechnung schrumpft.
What is a Translation Memory (TM)?
A Translation Memory is a linguistic database that breaks text into segments (usually sentences) and stores them as pairs of source and target language. When an identical or similar sentence comes up again, the system automatically proposes the stored translation. Unlike a glossary (which stores individual terms) or machine translation (which generates new), the TM works at sentence level and only retrieves previously approved translations. That makes it the most conservative and safest layer in a modern localization pipeline.
From 100 percent hits to fuzzy matching
/ OUR APPROACH
01
Step 01
Segmentation
The system breaks the source text into segments. By default a segment is a sentence, optionally a paragraph or a table cell. Every segment is enriched with metadata: which document it came from, which product line it belongs to, who approved it.
02
Step 02
Match check
Every new segment is checked against the TM. Identical sentences yield a 100 percent match (exact match) and are reused without effort. Similar sentences yield a fuzzy match between 75 and 99 percent, and a translator only edits the difference. Example: "Press the red button" vs. "Press the blue button" yields 90 percent similarity. The translator corrects one word.
03
Step 03
TM, MT, and glossary combined
The pipeline checks in this order: first the TM (the safest, approved translation), then the glossary (brand terms like product names, corporate terminology), only then machine translation for anything not yet in the TM. You get maximum reuse of approved content at minimum risk.
04
Step 04
TMX import and no vendor lock-in
Your TM belongs to you. We import existing TMs from SDL Trados, MemoQ, or Phrase through the open TMX standard (Translation Memory eXchange). Moving in does not waste prior investments. Moving out is always possible through a TMX export. No vendor lock-in, no data prison.
A memory that does not forget
If translator A in January translated "Button" as "key" and translator B in March independently chose "button," your customer is confused. The TM forces B onto the already approved version. Consistency across years and teams without anyone needing to maintain a glossary. For brand-critical terms (product names, claims), the TM is supplemented with a term-base lock: certain terms must not be changed.
40, 60 %
lower translation cost on recurring documentation
Industry value for existing clients in technical environments after TMS rollout. On manuals, technical datasheets, and product catalogs, 100 percent match rates of 30 to 50 percent are realistic. On creative marketing content (headlines, slogans), match rates are significantly lower. That is where MTPE is the right layer, not TMS.
How word-rate billing changes with TMS
Match type
Meaning
Billing (example)
100 % Match (Exact)
Sentence is identical in memory
0 to 10 % of word rate
Fuzzy Match (95-99 %)
Practically identical, small correction
20 to 30 % of word rate
Fuzzy Match (75-94 %)
Sentence is similar (e.g., only number/color differs)
30 to 60 % of word rate
No Match (New)
Sentence is entirely new
100 % of word rate (AI or human)
Tech stack: TMX as the open standard
We use the open TMX standard (Translation Memory eXchange) for importing and exporting your TMs. Existing investments in SDL Trados, MemoQ, or Phrase can be brought in. A future move away from us is always possible. Your translations are your company asset, not our data monopoly.
Three questions localization owners ask
You do. The TM is your company asset. We host and maintain it technically, but the data is yours. You can export it as TMX at any time and use it elsewhere. At the end of a contract you automatically receive a full export.
Conditionally. Marketing copy (slogans, headlines, hero copy) rarely repeats and often needs creative transcreation. Full Post-Editing (MTPE) is the right layer here. For catalogs, datasheets, and recurring newsletter blocks, TM is gold.
Through alignment. We take your existing translated PDFs, web pages, or Word documents and align them segment by segment against the source. The result is an initial TM that delivers match rates from day one. You do not start from zero, you build on your existing inventory.
Related deep dives
Market Localization and SEO
What TM does for words, localization does for markets: date formats, currencies, cultural codes, and keyword volume in the target market.
Quality Assurance for Translations
How automated QA checks (LQA, Linguistic Quality Assurance) secure consistency and tone of voice after the TM pass.
// TRANSLATION MEMORY
Let's review your savings potential.
Give us a sample document from your documentation or catalog. We analyze the repetition rate, estimate the match profile, and tell you concretely how much a TMS would save at your volume. Pseudonymized, in a week.