A large chunk of the annual report you send me in the fall is a report I already translated the year before. Accounting policies, boilerplate note headers, segment descriptions, governance tables—they barely move. A translation memory is how I bill you less for that repetition and, more importantly, keep every IFRS term identical to how it read last year.
Why is a quarter of my Geschäftsbericht last year’s Geschäftsbericht?
Because annual reports are structurally repetitive. As one localization vendor puts it plainly, “it’s likely that your annual reports are filled with repetitive content year-over-year.” The repetition clusters in predictable places: the standard IFRS and HGB accounting-policy notes, the governance and remuneration tables with their fixed labels and fixed structure, the recurring risk disclosures. These sections carry meaning forward almost verbatim; only the figures inside them change.
In my own annual-report projects, an MDAX report entering its third consecutive cycle typically hits 30–40% exact-match coverage against the prior-year memory. Those segments arrive pre-verified for IFRS terminology, carry a reduced per-line rate, and need editorial review rather than fresh translation. That is the cost opportunity everyone talks about. The consistency risk is the half nobody mentions: if last year’s “Grundsatz der Wesentlichkeit” became “materiality principle” in note 3 and “principle of materiality” in note 9, a memory is what stops that drift from compounding across cycles.
What is a CAT tool, and is it the same as machine translation?
No. A CAT tool—computer-assisted translation—is a professional environment that a human translator works inside. It stores every approved sentence pair, enforces terminology, and handles formatting so I never re-invent the same IFRS phrase twice. As Smartling describes it, a CAT tool “aids professional translators in performing their jobs” and “ultimately, the quality of the outcome relies on their linguistic skills.” Machine translation, by contrast, “doesn’t require human input”—it produces raw output on its own.
That distinction matters when you read a quote. The tools I use (Trados Studio being the industry standard) automate the repetitive mechanics so “translators don’t have to repeat translations or struggle with formatting.” They do not translate for me. The translation memory that lives inside the CAT tool is an assist to a human decision, not a substitute for one. If you want the separate case for where machine translation plus post-editing genuinely earns its keep on templated content, that’s a different service line with its own rate.
How does a translation memory decide what I pay for each segment?
A translation memory sorts every source segment into match buckets, and the bucket sets the rate. There are three you need to understand:
- Exact (100%) matches—a full repetition of a segment already in the memory, approved last cycle. These are the cheapest tier and the fastest: I confirm the segment still fits its context rather than translating it.
- Fuzzy matches (50–99%)—segments that resemble a stored one but differ. A 95–99% match is nearly identical (often just a changed figure or date); a 50–74% match shares structure but needs real work. As LocAtHeart notes, “matches ranging from 1 to 99% are called fuzzy matches,” and “matches below 50% are considered useless”—they make translation neither better nor faster, so anything under 50% is billed as new.
- New words (no match)—the genuinely fresh content: this year’s CEO letter, revised MD&A narrative, new CSRD disclosures. Full rate, full time.
There’s also a fourth, quietly valuable case: the context match (sometimes labeled 101%), where a segment is identical and sits in identical surrounding context. In a recurring annual report, that’s the tier that carries whole boilerplate paragraphs straight through.
My CAT grid, per standard line of 55 characters including spaces, works out as follows. A standard line is the German-market billing unit; if you buy translation in Germany, this is the number your quote is built on.
- Context match: €0.00 per line
- Repetitions / 100% match: €0.17
- 95–99% / 85–94% / 75–84% fuzzy: €0.51
- 50–74% fuzzy: €0.68
- No match (new): €0.85
That grid is the mechanism behind the savings. The industry rule of thumb bills exact matches and repetitions at roughly 30% of the new-word rate and fuzzy matches at about 70%, “because a fuzzy match will take a translator more time than a full match to review.” My grid follows the same logic. On a report where roughly half the content is repeats and fuzzies, that can approach 40% off the full-price total, though I’d rather quote your actual file than promise a percentage in advance—every report’s mix is different, and the €20 minimum charge still applies to small jobs.
Why do IFRS and HGB reports demand this kind of terminology discipline?
Because in a financial statement, terminology is not style—it’s meaning. The standard for this work is not general readability but equivalence for a qualified accounting reader: as one buyer’s guide frames it, every “figure, label and accounting term carries the same meaning a qualified reader would take from the original.” Treat the multilingual annual report as a regulated product, not a marketing translation.
German accounting terms don’t map one-to-one onto their IFRS English equivalents, which is exactly where inconsistency creeps in. A “Rückstellung” under HGB “doesn’t directly map to a ‘provision’ under IAS 37, because the recognition criteria differ materially.” Now imagine “Grundsatz der Wesentlichkeit” rendered three different ways across six notes. Each variant is defensible in isolation; together they read, to an auditor or a group reviewer, like three different concepts. As one financial-translation firm warns, “even minor discrepancies—a misinterpreted IFRS term, inconsistent terminology or a transposed digit—can alter the meaning of an entire report.”
A translation memory removes that risk at the source. Once the approved rendering of a term is stored, every future occurrence pulls from the same decision. Terminology fights sink timelines otherwise—decide the high-impact terms early, and the memory keeps them decided.
What does the Terminologiedatenbank do that the memory doesn’t?
The translation memory stores sentence pairs; the terminology database (Terminologiedatenbank) stores approved term pairs and enforces them everywhere, even inside brand-new sentences the memory has never seen. The two work together: the memory ensures “consistent terminology across all reports,” while the term base ensures “all official terms, abbreviations and spellings are automatically used correctly” in line with IFRS, HGB, ESRS and CSRD.
For recurring clients I maintain a term base that captures company-specific terms, management’s preferred phrasings, and the historical decisions that keep year-over-year consistency intact. A compact bilingual sample of what those locked entries look like:
German (source) English (approved) Note
────────────────────────────────────────────────────────────────────────────
Rückstellung provision IAS 37 — recognition
criteria differ from HGB;
flag on first use
Pensionsverpflichtungen defined-benefit obligations IAS 19
wesentliche Tochtergesellschaften material subsidiaries consolidation scope
Grundsatz der Wesentlichkeit materiality principle one rendering, lockedEach pair is decided once and reused, so the same concept never appears under two labels across the notes. That is the plain-English purpose of the term base: defined-term decisions captured once, then propagated and locked.
Who owns the translation memory, and what happens if I change providers?
You do. The memory is a record of approved terminology that you paid for—not a byproduct the vendor gets to keep. If it’s locked inside a vendor’s system with no export provision, you “may lose access to work you already funded.” That’s an avoidable trap, and it’s why I hand clients their TM file at the end of every project.
Portability runs on an open standard called TMX (Translation Memory eXchange), “an XML specification for the exchange of translation memory data between computer-aided translation and localization tools with little or no loss of critical data.” TMX has long been the vendor-neutral format, and version 1.4b is universally supported. Export your memory as TMX and it moves into any other CAT tool. One soft caveat: some tool-specific metadata (certain Trados context fields, for instance) may not fully round-trip, so “little or no loss” is honest wording rather than “zero loss.” The practical guidance is simple—request regular TM exports, review the asset periodically, and work with someone explicit about client ownership. If you switch providers mid-cycle, a clean TMX means the new translator starts with your history intact instead of from scratch.
Trados or Across—does my LSP’s tool choice affect my schedule?
It can, and it’s worth knowing before production starts. Many listed companies run their translation supply chain on the Across Language Server, a server-based platform built as a “closed system” that “ensures security for translation projects and automates the processes along the entire supply chain.” Its central translation memory and integrated terminology system live on the LSP’s or client’s Across server, not in a file that ships back and forth.
That closed architecture has a concrete production consequence: an external translator can’t simply open your job in Trados. They have to work inside the Across ecosystem—the Translator Edition or the web connector—which means your project needs a translator who is Across-certified, not merely Trados-fluent. This is a workflow-architecture difference, not a measured delay, but if your LSP is on Across and your translator isn’t set up for it, that mismatch surfaces exactly when your reporting deadline has no slack. I work in both environments; the point is to confirm the fit before the contract, not during the busy season.
What should I ask my translator before signing?
Three questions separate a vendor who treats your memory as your asset from one who treats it as leverage:
- Do you deliver the TM file at project close? The answer should be yes, in portable TMX, with no separate fee to release work you already paid for.
- Is the Terminologiedatenbank client-specific and locked against third-party projects? Your approved IFRS phrasings should never leak into a competitor’s report, and a rival’s terminology should never contaminate yours.
- Where is the TM stored, and is it covered by the NDA? Pre-publication financials are price-sensitive. The memory holds your unreleased numbers and narrative, so its storage location and confidentiality treatment belong inside the same NDA as the source files.
I answer all three the same way for every client: your file at close, your term base locked to you alone, your memory stored under the same confidentiality terms as everything else you send me. If you want to see how that maps onto a specific report, send me last year’s PDF and this year’s source—I’ll analyze the match distribution and quote against the grid above. Start at the contact page, or read the full rate structure on the services page.
One thing I don’t offer, for the record: certified or sworn translation. If your filing needs a court-sworn attestation, that’s a separate credential and I’ll point you to colleagues who hold it. Everything above is about specialist financial translation—getting the terminology right and keeping it right, cycle after cycle.