A translation memory, or TM for short, is a database of text segments in a source language (L1) and their translations in target languages (L2 ... Ln). To start a translation with a translation memory, the translator instructs the TM to cut the text into segments. These segments are usually the size of a sentence or part of a sentence and can be adjusted by the translator.
For every segment, the TM will search for an exactly matching segment in the database. If no exact match is found, the TM can search for approximate matches. The matches are presented to the translator for editing and approval. If no match is found, the translator has to translate the segment after all. When the translator approves the translation, the TM stores the original segment alongside the translated segment in the database.
It is advantageous to use a TM if translating manuals[?] for products that have many identical sections, or if a manual for a product is based on an older, already translated manual; in those cases, a TM can save a translator a lot of time. However, using a TM for literary translations is not advisable, simply because these is so little repetition in these works.
Commercial software
wikipedia.org dumped 2003-03-17 with terodump