How does a Translation Memory work?

We have mentioned quite a lot in the past the use of Translation Memories as a means to obtain a consistent yet cost efficient translation at little to no effort from our translators. What we want to do now is to explain to you why and how this works, allowing you to better understand our system.

Using a translation memory does not mean using a machine translation but rather re-using previous translations. The Translation Memory, or TM, represents a database of „pairs”, formed by sections of already translated and approved text, in both source and target languages. Usually, the software we use in this regard (whether it is alignment programs or SDL Trados), divides the segments automatically, usually in sentences, list or table elements.

Any new text is then compared to the one stored in the database and similar segments will be recognized and shown as a ”fuzzy” match. When composing a project package, a word count analysis will be generated and contain the following:

  • Perfect Matches (100% match): This segment in the new document is identical to an already stored one contained by the Translation Memory.
  • Repetitions: This particular segment is repeated multiple times in the new document and has to be translated only once, the rest being auto-generated by the TM throughout the project.
  • Fuzzy Matches (75-99% match): This segment is partially identical to one in the TM but not entirely, needing revision.
  • No Matches (0-74% match): No similar segment has been found and the text must be translated entirely.

During the translation process, the system shows what percentage corresponds to each segment and allows the translator to accept this suggestion, to modify it or to replace it entirely with a new translation. These new additions shall be stored in the Translation Memory to use in future projects.

A translation memory can be either created using an empty one, used to store the translations from the beginning of the project, or it can be created from existing translations, using an ”alignment” tool. Such tools automatically line and pair a source file with the approved, translated one and transform them in databases for future use.

Using a translation memory system has the following advantages:

  • The entire translation process is much faster, being most efficient for things like updates or documentation for software, where the volume of new translations is much reduced;
  • It allows for an easier method of tracking and replacing the parts that need to suffer changes;
  • The new translations are much more consistent, because the same terminology can be used throughout the projects of the same client and it can also allow integration of glossaries;
  • It is cost efficient, since the word count that has to be translated from scratch is reduced and it allows implementing a series of discounts.

Our quotation discounts:

Perfect Matches or Repetitions:

Since the text to be translated is identical to the previously existing one, the only thing remaining is checking the integrity of these segments and only adjust the terminology when necessary. Normally, these are discounted between 75% and 100%, on a project-to-project basis.

Fuzzy Matches:

For the 75-94% matches, the previously translated text can be reused, with only minor edits and adaptations, and we apply a discount of 25%. This price includes the review, after translation.

No Matches:

If the match is lower than 75%, the text will be translated entirely, which means the full price is applied, including of course, the linguistic review.

Now, knowing these, we hope you can understand what managing a translation or localization process entails and you can check more of our services!

If you like this post, please share!

Comments (1)

  1. Reply

    Normally, memoQ uses the source text from the previous and the next segments as the context. In most cases, this is the preceding and following sentences for the current sentence (101% match). However, memoQ can handle context this way only if the document contains running text. The way memoQ handles context depends on the document format. The most prominent running-text formats are Microsoft Word documents, HTML, plain text documents, FrameMaker and InDesign documents. For tables (such as Excel workbooks) and data structures (XML files or software resources), the context is defined in a different method (ID based). For tables, you can choose another column or another cell to serve as context for each cell to be translated. You can set this up in the appropriate Document import settings window (in this case, the one for Excel). For XML files, you can use either neighboring elements or attributes to serve as context. You can set this up in detail in the Document import settings window for XML files .

Leave a comment

Your email address will not be published. Required fields are marked *