Translation memory is a database containing pairs of translation units (segments) of the original text (source segments) and their translations (target segments). CAT applications use these databases for fuzzy translation look-ups to simplify and speed up the translation process, eliminating the need to translate text that had already been translated before.
Types of Translation Memories
Depending on their physical location, MemoQ recognizes two types of translation memories:
To be able to use a translation memory, you need to add it to your project (use project-level translation memory management or global-level translation memory management).
Creating Translation Memories
There are several ways to create translation memories:
1. When you use MemoQ and have at least one translation memory added, the program stores segments you translate to this memory. If multiple translation memories were added to the project, the primary translation memory is used.
2. If you have previously used another CAT tool, you can export your translation memories from the application you are migrating to MemoQ. Translation memories can be exported to a TMX file and then imported in MemoQ.
3. You can also add entries into the translation memory by using the alignment function or create them manually by means of direct editing of the translation memory.
Translation memories help you recycle previous translations, thus increasing your productivity. During translation, translation segments are compared against segments stored in the translation memories added to your project. When this comparison produces a hit, i.e. when a certain level of similarity is found, results are automatically displayed, so that you can see how you (or another translator) has previously translated similar or identical segments.
Depending on the level of similarity of two segments (also called hit rate), hits are divided into several categories:
Segments obtained by means of alignment or negation can also be penalized, i.e. a specific percentage (penalty) can be defined, which is automatically subtracted from the given segment's hit rate to indicate that such segment was not created manually by a translator.