Whereas books traditionally comprised the bulk of translation work, the focus has since shifted to digital media. These ‘new’ media include websites, mobile applications and games. Not only is the content different, so is the approach taken with respect to translation. In the 19th century the typewriter drastically changed the rules of play, and today, software applications are causing a similar revolution. These translation tools use technology such as machine translation (MT), terminology base (TB) and translation memory (TM).
What is translation memory?
Essentially, translation memory is simply a database that stores text segments. A translation of a specific text can be added to the database per language. The translation tool detects the text segments by means of segmentation. Each segment consists of one or more words or even sentences. Should a text segment and its translation into the desired language be available in the database, this segment can be translated by means of TM. When a text segment appears multiple times in a text, the repetitions are translated all at once. The translation tool will recognise the segment throughout the text. Entering and confirming the translation of a text segment in the database for one instance will automatically translate the same segments in the entire text.
Slight variations between a text segment and a segment from the memory will also be recognised. For instance, if a text segment is the same as that in the TM, except for a difference in the verb tense, the TM will still recognise the text. A segment with a lower similarity percentage will prompt the TM to offer the same translation suggestion, unless a better option is available. This yields more consistent translations, especially for automatic translations.
Automated translation is often associated with the use of machine translation. However, translation memory can also be used. Some translation tools have a ‘pre-translation’ feature that allows you to specify a threshold. All segments that can be translated by TM and have a recognition percentage above that minimum value will be translated automatically.
The major drawback to using a translation memory is that the TM database must first be constructed. Building this TM can be a long process and does not guarantee a better translation. For instance, translations geared at specific target groups can go horribly wrong. Using a TM for extremely specialised topics can also be difficult, due to the rare nature of the subject matter. Nevertheless, TM is invariably useful for translating texts with a high number of repetitions. Consistency and automatic translation of all identical segments will always be made more efficient and consistent by using TM, irrespective of the size of the overall text.