On Translation Memories and Terminology management
Translation memory systems make translations more consistent and cost-effective. But how can these systems help standardize terminology? This blog post offers an answer from the perspective of an experienced translator and project manager.
The translation agency you rely on is always good for a surprise. You’ve been working together for years and there are plenty of entries in the translation memory, but suddenly, in the last translation you requested, the English term “countersunk screw” was translated into German as “Senkschraube.” Until now, this term had always been translated as “Senkkopfschraube.” You can easily confirm this fact in the included translation memory. How could this happen? And more importantly, how can you systematically prevent this type of error?
To answer this question, it makes sense to take a quick look at how translation memory systems work. As a translation database, a translation memory is primarily a means of rationalizing translation. The translations you ordered in the past are archived there in the form of sentence pairs.
Brought together by the translator: the sentence pair
The term “sentence pair” requires some explanation. During translation with a translation memory system, each source text is divided into segments or translation units based on certain segmentation rules. A translation unit consists of the segment in the source language and the segment the translator just wrote in the target language. These two components are stored as a pair in the translation memory.
Generally, segmentation into translation units is based on sentence and paragraph separators, meaning periods, exclamation points, question marks, colons and paragraph marks. Paragraph marks, for example, allow separate individual words or items in a list to be saved as individual segments in the translation memory. Strictly speaking, you would actually have to call them “segment pairs.” But the typical segment is still a sentence.
The translation memory – a database to jog your memory
For every sentence (or rather segment) that has to be translated, the system will automatically search the translation memory like an archive to see whether a similar or identical sentence is already saved in the target language there. If a matching sentence is available, the translation will immediately appear as a suggestion. Any differences between the current source-language sentence and the source-language sentence found in the translation memory are highlighted.
The translator saves time, and you, the customer, save money. This solution also improves consistency because identical sentences are translated the same way, and similar sentences are translated in a similar way.
In addition, the translator can use a concordance search function to find out whether his or her colleagues have already translated a certain word or phrase, and if so, how they translated it. However, the translator will only use this function if he or she thinks that the word might have been translated before. He or she also has to realize how important it is to look up the word, for example because the word is specifically defined in the terminology. Terminological consistency can only be ensured if the translator actively uses the concordance search, finds a match and applies it to the translation.
In the case of the countersunk screw, the translator either didn’t use the concordance search, or deliberately decided to use a different term, for example because he or she mistakenly thought that “Senkschraube” was the more commonly used term.
How can you best avoid this type of deviation in terminology?
Simply instructing the translators to use the translation memory as a mandatory terminology source can narrow down the translator’s terminology choices from the very start of the project. This way, the translation memory no longer just describes the status quo, but also defines the standard for all terminology questions. Unless a separate list of terminology is available, this is also the general practice. For this reason, it’s unlikely that the translator made a conscious decision about terminology in our example.
But the translator would be more likely to use the correct term if he or she had a list of customer-approved terminology in addition to the translation memory. This list would make it clear that the English term “countersunk screw” has to be translated as “Senkkopfschraube.”
In order to avoid misunderstandings, this list of terminology should contain not only terms in the source and target languages, but also definitions of the terms. Definitions help translators know exactly when to use a term. An example: The English word “file” can mean either a file in an archive, which would be translated into German as “Akte,” or a computer file, in which case it should be translated as “Datei.” With a clear definition, the translator is able to determine the right target term for a given context.
Of course, even with this type of list or database, the translator still has to consciously and actively look up terms. But, unlike the translation memory, it can also guide future translations. In other words, a terminology list or database can contain terms or designations that aren’t in the translation memory or don’t appear very often, but that will be very important for documents and topics that have not yet been translated.
To take it a step further, couldn’t the translator receive a notification whenever he or she is translating a sentence or segment that contains a word from the terminology database? Modern translation memory systems make that possible. The relevant terminology database is integrated right into the processing environment and combined with term recognition. For example, if the translator is working in SDL Trados Studio with an integrated terminology database, he or she can immediately see that there is an entry in the terminology database for “countersunk screw,” and the correct translation appears as a suggestion.
Terminology management within the translation process
Of course, this only works if an appropriate terminology database is available. At the same time, putting together a set of terminology takes a certain amount of work, even if you only enter the data that is absolutely necessary. In some cases, certain steps in terminology work (such as recording terms) can be incorporated into the translation process. But this is not the most reliable way to handle the terms (read about a systematic approach to terminology work here), nor is systematically recording and protocolling terminology decisions the main focus of translation. On the contrary, separately recording terminology clearly represents additional work that the translator can’t do for free.
Whether before or during translation, it pays to take a methodical approach to terminology work and use terminology databases. After all, terminology databases do things that translation memory systems can’t: They set standards, not only describing the status quo, but also laying down guidelines for texts and topics that haven’t been translated yet, and – through direct integration into the translation memory system – they can help translators to systematically avoid terminology errors like “Senkschraube” instead of “Senkkopfschraube.”
Go ahead and ask the project managers or language experts at your translation agency where and how coordinated terminology could help you. We would be happy to advise you.
About the author
Frank Münnich has been a translator and project manager at text&form since 2005. He currently specializes in translating marketing texts. He studied English and American Studies, as well as Germanic Linguistics, at Humboldt Universität in Berlin, where he earned his Master of Arts degree. From early 2007 until mid-2008, he was in charge of maintaining and enhancing extensive terminology databases.
