Translation memory is a bit of translation industry jargon that is frequently misunderstood by customers and by software developers. This tutorial explains how translation memory works, and how it can be used to maximize quality while reducing translation costs.

Translation memory or TM is essentially a database of human edited translations that is used to reduce repetition, improve translation consistency and provide translators with the ability to view similar texts and their translations (to improve style and consistency). It is an essential translation automation tool, and is incorporated into most CAT (computer aided translation) utilities and systems. However, there are several different types of translation memory, each of which is used differently and provides different advantages to the user.

Exact Match TM

The simplest type of translation memory looks for exact matches on strings or blocks of texts. From a software development and computing resource perspective, this type of query is fast and scales well for large data sets. Its primary use is to look for texts that have already been translated and re-use the translations, both for consistency, and to avoid doing duplicate work.

This type of translation memory does not save much money for long form content, such as this blog, because almost every sentence on every page is unique. On the other hand, it is very useful for large data sets, such as a database of product descriptions, or a message catalog for a mobile application. These types of systems tend to have a lot of short texts with a large number of repeats. Exact match TM will pick these up, so they are not continually re-translated.

Virtually all computer aided translation systems have this capability. If a system doesn’t do this, it is a serious deficiency that should disqualify it from consideration.

Term Glossary

Term glossary is another type of translation memory that is used to insure that certain terms and phrases are always translated consistently. In a typical CAT tool, the translator’s editing tool will highlight words and phrases that appear in the term glossary and their recommended translation. The translator can then accept or modify the recommended translations. As a translation buyer, you’ll be asked to prepare a dictionary of terms and phrases you want to be included in your term glossary.

Term glossaries are not used so much for cost saving as they are to insure consistency, especially when translations are being done in parallel by a large number of translators. This is particularly important in protecting brand names, translating technical terms, and enforcing a common house style.

Most CAT tools support term glossary, although they differ in how the utilize them. Some systems will auto-translate term glossary matches, but the best practice is to allow the translator to make the final decision to accept or override these suggestions (for example if the recommended translation is in the wrong tense, or needs to be edited for grammar).

Fuzzy Match TM

Lastly, fuzzy match TM is used to pull up similar long form translations (typically full sentences). This type of translation memory is not used to automatically substitute or translate texts, but to provide the translators with a popup view of similar texts and their translations. The translator decides whether to use a translation, or part of a translation. The translator can also passively read the previous translations, and simply use those as a style guide.

This type of translation memory is most useful in translating long form content. It is not generally used for cost savings because few texts will be directly usable. Like term glossary, fuzzy match TM guides the translators to provide consistent style, and in learning the subject matter they are translating. Fuzzy match TM is available in the majority of CAT tools but is not universal (from a technical perspective, it is actually quite difficult to do very well).

A Word About Cost Savings and Perverse Incentives

A common myth among customers is that translation memory can produce dramatic cost savings. However, apart from preventing the re-translation of exactly matching texts, translation memory is best viewed as a quality assurance tool. There are other techniques that you can use to reduce translation costs, by an order of magnitude, which I will describe in an upcoming tutorial.

It is actually best if the translators are paid equally regardless of whether they use a translation from a term glossary or fuzzy match TM. Why? Because if the translator is punished for using the translation memory, guess what will happen? They won’t use it.

Summary

Most translation services and tools support one or more types of translation memory. For you as a translation buyer, you won’t need to worry about how these tools work apart from preparing a dictionary for your term glossary if you need this.