Translation memory explained: stop translating the same sentence twice

If you translate for the same clients again and again, you translate many of the same sentences again and again. Contract clauses, safety warnings, product descriptions and email signatures come back in every new document, often with only a word or a number changed. A translation memory, usually shortened to TM, exists so that you never have to translate those sentences from scratch again.

What a translation memory is

A translation memory is a database of sentence pairs: a source sentence and the translation you approved for it. Every time you confirm a sentence in the translation editor, the pair is stored. When the same or a similar sentence appears in a new document, the memory finds it and offers your earlier translation.

It does not translate anything by itself. It remembers your decisions, so that the same sentence is always translated the same way, by you or by anyone on your team.

Exact matches and fuzzy matches

• An exact match (100%) is a sentence that is identical to one already in the memory. The stored translation can be inserted directly.
• A fuzzy match is a sentence that is similar but not identical, for example a clause where only the number of days has changed. The editor offers the earlier translation, and you adjust only what changed.
• No match means the sentence is new. It is translated by the AI and reviewed by you, and once confirmed it becomes part of the memory too.

The more you work for a client, the more of each new document is already covered by the memory.

Why it matters

Consistency

Clients notice when the same clause is translated in three different ways in three documents. With a memory, the approved wording comes back every time. This is especially important for legal, medical and technical texts, where a small change of wording can change the meaning.

Speed

Repeated sentences are filled in at once, so you spend your time on the sentences that are actually new. Updated versions of a document, a contract with a few changed clauses or a manual with a new chapter, are where the memory saves the most time.

Quality over time

Every correction you make is stored. The memory gradually becomes a record of your best decisions for each client.

How to keep a memory clean

A memory is only as good as what you put into it. A few habits make a big difference:

• Confirm only sentences you are happy with. Whatever you confirm will be offered again.
• Keep separate memories for clients or subjects that use different terminology. A translation that is right for one client can be wrong for another.
• If a wrong translation got into the memory, remove it and confirm the corrected sentence. Otherwise the same error comes back next time.
• Combine the memory with a glossary. The memory stores whole sentences; the glossary fixes individual terms, so new sentences use the right words too.
• Remove duplicates where the same sentence has two different translations, unless both are needed for different contexts.

A worked example

Suppose you translate the terms and conditions of an online shop. In January the shop sends you version 1, about 3,000 words. You translate it, review it and confirm every sentence, so all of it goes into the memory.

In June the shop sends version 2. Most clauses are unchanged; some have a new number of days for returns; two sections are completely new. When you open version 2:

• the unchanged clauses are exact matches and are filled in with your approved January translation;
• the clauses with a new number are fuzzy matches, so you only change the number;
• the two new sections are translated by the AI and reviewed by you, then confirmed into the memory.

Instead of reading 3,000 words again, you spend your time on the few hundred that actually changed, and the client gets a version 2 that uses exactly the same wording as version 1 wherever the text is the same.

Memory, glossary and AI: who does what

It helps to see the three tools as a team. The translation memory remembers whole sentences you have already approved. The glossary fixes individual terms, so that a product name or a legal term is always translated the same way, even in new sentences. The AI translates everything that is new, following the glossary. Your review is what turns the AI output into approved translations, and those, in turn, feed the memory for next time.

Bringing your existing memories

If you already work with a CAT tool, you probably have memories in TMX format. DeepReference imports TMX files, as well as glossaries in TBX and CSV and bilingual XLIFF files from other tools. Rows that are exact copies are skipped during import, while a sentence with a different translation is kept, so you can decide which one is right.

Who it is for

Translation memory is included in the Translator Pro and Business plans. In Business, a team can share memories and glossaries, so everyone working for the same client uses the same approved translations.

In short

A translation memory is the simplest way to make your translations faster and more consistent at the same time. Confirm carefully, keep memories separate where terminology differs, fix errors at the source, and let the memory do the repetition for you.