How Translation Memory Works

Get the most from your translation assets.

Translation memory systems first analyse all your files’ text contents. It will review each sentence and process your files as follows for the translator. Translation Memory is often referred to as TM or computer-assisted translation.

Real humans actually do the translation and the TM system helps them by showing them previously translated texts.

In Action

Firstly, the translation memory system will check for reusable text from previously translated files, pages and paragraphs. It looks to see if you have translated anything before.

Then it will review your documents and look for similar sentences (fuzzy matches). These are then presented to the translator in the same way as predictive text works on your phone, so they are use and modify them accordingly.

How does translation memory work?

The Process

Let’s examine the process of translation memory in more detail.

Reusing text from previously translated files, pages and paragraphs

First, all the text in the file is compared to all the text of other previously translated files. If a file is found where all the text matches then all the text from the matching file is reused as a complete file. This text will be counted as a what is called context match. It will not be proof read or rechecked. This process is repeated for the smaller units of complete pages and then repeated again for complete paragraphs.

Reusing text from previously translated sentences

Supposing we have a paragraph with 3 sentences, where the first sentence and the last sentence matches previously translated sentences, but the middle sentence is completely new. In this case, the first and last sentences will be counted as perfect matches and the middle sentence will be counted as new words.

We will translate the middle sentence from scratch. We will check that the translation for the first and last sentence is still appropriate, given that we now have a new middle sentence. Checking that a sentence is still appropriate takes a lot less time than translating it from scratch. As it takes less time and so it costs less, it is charged only at the perfect match rate.

Reusing text from sentences that are similar to previously translated sentences

It will take less time to translate a sentence that is similar to another sentence that was previously translated than it will to translate it from scratch. However it will it takes more time to translate a sentence that is similar to another previously translated sentence than checking a perfect match sentence. Translating sentences that are similar to other previously translated sentences are charged at the fuzzy match rate.

Translating repeated text

Sentences that are identical to other sentences are called repeated sentences. Repeat sentences only need to be translated once. Each repeat sentence is checked in the location it is used. This is similar to processing the perfect match sentences and so is only charged at the perfect match rate because it takes less time than translating it from scratch.

Read more about translation memory on Transit NXT: Translation Memory software.