Whenever a translation project is started a good part of the required translation already exists. It comes in form of reference material, but not as a ready to use memory which can be used in a translation memory system (eg as a pdf file). Which of course is not the best thing to start with. On one hand there are additional expenses for the same translation. Also, no translator is willing to check terminology or individual sequences in two different reference documents (original and translation) for consistency. So it is no surprise that already translated sentences or terms are used in an different way in the new translation project.
For such cases, the so-called alignment may bring some relief. What happens during an alignment? In theory, this process is done in three stages. First the original source text and the translation are read by an alignment program or an alignment module within a translation memory system (TMS). These sentences are then decomposed into individual segments. A segment is a (configurable) translation unit, usually in form of a sentence. In a further step the alignment program, using internal algorithms, assigns the segments to the respective languages. In a last step, a language expert checks whether the sentence pairs fit together correctly.