translation memory

The Difference Between Translation Memory and Machine Translation

Often, when we run webinars that feature a demonstration of SDL Trados Studio 2011, we get asked to explain the differences between translation memory and machine translation as tools to help translator productivity. As this has proven to be a popular topic, this blog will provide a high level overview of the difference between translation memory and machine translation from the point of view of someone accessing these resources through SDL Trados Studio.

Translation Memory

A translation memory (TM) is a linguistic database that continually captures your translations as you work for future use, but how does it actually work in translation software? SDL Trados Studio 2011 will break down a source file into segments. A segment is a manageable bite sized chunks. As these source segments are translated, they are saved to the TM. At the same time segments are being saved for new translations, the TM is also being used to leverage previously translated content. When you move to a new segment for translation, the software checks in the TM if there is an identical or similar translation and automatically enters the result which is most appropriate into the new target. Any match with the TM is given a percentage score depending on how accurate it is. In Studio 2011 a translator could get a PerfectMatch, a ContextMatch or a 100% match, all of which can be used as it is. Studio can also display a ‘fuzzy match’, which is under 100% and may require some post editing. Either way, significant amounts of time are being saved across the board when TM is being used.

So the TM content comes from a human translator, opposed to machine translation which is being generated by a computer. If you want understand more about TM a great place to start would be SDL’s monthly webinar titled ‘What is Translation Memory’. Visit our webinar calendar to reserve your place. Machine Translation

Now, onto machine translation (MT). To put it simply MT is the process of changing text from one language into another language using a computer, i.e. without the input of a human translator. When using Studio 2011, both translation resources populate the target segment in the same way but unlike with TM, untrained MT does not provide you with a match percentage for each translated segment, so it relies on the translator or reviewer to judge how accurate the suggested translation is. The quality of translations can vary significantly, and sometimes the results provided by machine translation can be quite amusing.

Of course all machine translation solutions were not made equal and it ranges in sophistication between rule based MT where linguistic rules are applied to bilingual dictionaries, to statistical based MT when a calculation is made based on syntax to provide the translation. Statistical based machine translation is generally considered to be more accurate and the more sophisticated machine translation solutions work in this way. There is also untrained and trained MT. Trained MT solutions like SDL BeGlobal Enterprise are used by corporations to access trained translated content which is bespoke to their particular business. Untrained MT provides a generic translation and is more commonly used by translators. The best way to find out about MT is to give it a try. Studio 2011 makes it really easy to try out MT through offering Studio customers free access to SDL BeGlobal Community, SDL’s untrained statistical MT solution.

To recap, both TM and MT resources help translating content. A TM grows over time as you fill it with translated content, so the more you translate the more productive you become, whereas MT generates content immediately and needs to be retrained to see improvements over time. So it is where the translated content comes from and how this content is generated which is the key difference.