Translation memory software was first discussed in the 1970s, but it wasn’t until the 1990s that the first translation memory tools were made available to translators and customers. The main aim of any translation memory tool is to reduce the translation work needed by reusing as much of an existing text as possible. If you are not in the localisation industry, or familiar with certain concepts of information processing, then the inner workings of TM software might be alien to you. This article builds on the info given in our Ultimate Guide to Translation, and aims to give you a basic knowledge of how TM tools work and the benefits they can offer. Do you need to understand how translation memory software works? Perhaps you don’t understand how TM software works, but is this really a problem? As a general rule, not really. Your language service provider and their translators will be the main users of the translation memory tool, and you can safely expect them to be familiar with all of the advanced features that support their work. However, understanding the basic principles of translation memory software and how it considers units of language might affect the way you write your content. Writing with an awareness of potential pretranslation possibilities can lead to significant cost savings in the long term. What is translation memory software? Put simply, translation memory software stores “segments” of text that have been previously translated. It aids human translators by providing these existing translations as suggestions during the translation process. In so doing, it improves consistency of translation and speeds up the translation process, delivering cost savings. Basic principles of a translation memory software tool A TM tool considers a text as a series of language units, called segments. A segment can be a clause, sentence, paragraph or a sentence-like unit (headings, titles or elements in a list). Each segment within a source document is linked to a corresponding segment in the target text. Once translated, the translation memory software will store the translation for future use, either subsequently in the document or at a future date. At the start of any translation project, the translation memory software analyses the content and uses a fuzzy logic algorithm to consider exact matches and partial matches within the existing translation memory. What is a fuzzy logic algorithm? Generally, within computing, fuzzy logic considers “degrees of truth”, rather the “true or false” Boolean logic on which modern computers are based. When it comes to translation memory software, each segment is compared against existing translations to give a fuzzy match percentage. This percentage rating gives an indication of the similarity between the new text and an existing translation. The translation memory engine considers the words within the segment, looking at how many are the same, how many have changed, whether words have been added or removed, whether words have been moved within the sentence. It also looks at the capitalisation and punctuation of the sentence and takes this into account for the fuzzy match percentage. Again speaking in generalisations, the translation industry classifies a fuzzy rating of 70% or higher as a fuzzy match. Fuzzy ratings of less than 70% can have so many differences that it might take longer to amend the sentence than it would to start again fresh. What is the difference between translation memory software and a machine translation engine? The choice between a TM tool and a machine translation engine is something that we look at in other blog posts, but we can give you the basic overview here. Translation memory software uses reference material from existing materials to provide exact matches and suggestions of similar translations. The main work of translation is carried out by a human translator. A machine translation tool, also known as a machine translation engine, uses existing translations and artificial intelligence algorithms to translate the entire text. These translations can then be “post-edited” by a human to fix any errors made by the machine translation. What is an internal repetition? When looking at translation matches for a new document compared with an old document, translation memory software also looks at internal repetitions within the text. Internal repetitions are the subsequent occurrence of a sentence within your document. The first time a sentence appears, the translation memory software considers it as a new or fuzzy sentence, but for second (third, fourth, etc.) occurrences, the sentence is considered an internal repetition and is offered at a low rate. How does this benefit you as the customer? At a first glance, it can seem that the choice to use translation memory software or not lies with the language service provider, and that it is independent of the customer. Often, the customer does not run their own file imports, nor do they manage their own reference material. However, this does not consider the whole of the issue. Although you might not directly interact with the translation memory software, you are directly affected by its use in terms of project cost and translation quality. How can translation memory software reduce your translation cost? As a first point, starting to use translation memory software for your localisation projects has the potential to dramatically reduce the cost of translation. As early as from the second translation order, you may start to see reductions in cost as the percentage of reuse from translation memory increases. These increases will vary depending on your text type; a technical manual will contain more crossover between documents than a series of marketing press releases. Additionally, however, with an awareness of translation memory software and how it works, content creation that is optimising your texts for translation reuse can further increase the likelihood of pretranslated matches within their documentation. How can translation memory software decrease the turnaround times for your translations? As well has having a positive effect on the cost of translation, use of translation memory software can also speed up translation. For our in-house team, we assume that one translator can work on 2000 words per day in a completely new document. If the same translator is reviewing a pretranslated document, we would estimate that they can review 10,000 words per day. That’s five times faster! These are conservative estimates, so actual speed increases could be even higher. How can translation memory software improve the quality of your translations? Translation quality is notoriously difficult to measure as it can be highly subjective. For the purposes of this article, we will look at two factors in translation quality that can be definitively linked back to the use of translation memory software. The first of these factors is consistency of terminology. TM tools includes a terminology module that can be used to provide approved terminology to the translator. During the quality assurance phase of the project, this terminology module can automatically check that the approved terminology has been implemented. The second factor wherein translation memory software can support translation quality is using integrated QA checks. There are many different quality assurance checks available to translation suppliers. In addition to the standard spellcheck that you might expect, there is also a format check. This is particularly useful for checking the correct usage of punctuation, capitalisation and number localisation. For a long time, translation memory software has been the domain of the language service provider. Honestly, I’m not sure that I see this changing in the majority of cases. However, I do believe that knowledge is power. If you know a little bit more about the tools that are used to work on your projects, you can adapt your processes to make the best use of it.