By Matt Hauser

ComputerWorld.com – July 27, 2011

One of the most common things I see on a day-to-day basis when interacting with potential clients is confusion between machine translation and translation memory.  I recently covered machine translation, so in the interest of equal coverage, I will now focus on translation memory. 

A quick definition (for more information, check out the Wikipedia page):

Translation Memory (TM) is a tool that helps human translators to work more efficiently and with a higher degree of accuracy and quality.

So how does it work?  At a high level, translation memory creates a relationship between a segment of source language text and a corresponding segment of target language text. 

Here is an example:

You write the sentence, "My house is blue," on your company website and translate the phrase into Spanish.

"My house is blue" is now linked in the translation memory system with its target language equivalent, "Mi casa es de color de azul." 

Why anyone would have a blue house, or would want to publish this on their website, defies logic, but work with me here, please (note: blue houses are completely normal and this post is not intended to offend anyone who lives in one). 

The important thing is that the relationship between those two text segments is in place.  Why is this important?  For one, if that segment repeats itself across the site, it can be re-used automatically. So you are getting the benefit of accurate, human translation without having to pay for it more than once. 

Since the segment is being re-used, you also have the benefit of consistent language.  Language consistency is especially important to corporations for many reasons, ranging from maintaining brand voice in marketing content to increasing customer comprehension in informational content.  Language is extremely subjective, meaning that content can be written or expressed in multiple ways by different authors and have the same connotation or meaning to the intended audience.  The goal is to publish content that is consistent in the source language and then use translation memory tools to ensure that the translated equivalents are consistent, as well.   

Another benefit of re-using language is that it increases language accuracy. Each time the technology leverages a previously approved phrase from a database, it removes a human being from having to do a manual process. Therefore, using best-practice translation technology not only increases efficiency, but also increases language accuracy, because it mitigates the risk of introducing an error for segments which have been previously translated.

Since the gating factor in getting content to market is the overall number of words that need to be translated, by reducing the amount of work that needs to be put through a human process, you can go live much faster since you are eliminating manual, repetitive effort.

Another concept of translation memory is "fuzzy matching." This means that once your translation memory is created and updates are processed against it, the system can look for segments that are close matches (e.g.  "My house is red"), so that the translators just need to make minor modifications to the existing target language segment as opposed to an entirely new translation.

We will get into the benefits of server-based translation memory versus desktop-based translation memory in a future post, but the key thing to remember is that this solution offers multiple benefits to the overall translation process. 

So make sure that your vendor is using it, you're made aware of your savings from it, and whatever translation memory is created becomes your intellectual property.

Now I am off to paint my house blue...

Matt Hauser is VP of Technology for TransPerfect Translations.