Termbase and Translation Memory: What Is the Difference?

The two concepts most often confused in translation technology are the termbase and the translation memory. Both set out to reuse past work, but they store different things and work in different ways. Knowing the difference between a termbase and a translation memory is the first step towards understanding which tool answers which problem.

Contents

In short: a translation memory manages sentences, a termbase manages concepts. One answers the question “how did we translate this sentence before?”, the other the question “what term do we use for this concept?”. In this article we look at what each of them stores, how it works and how the two complete each other.

What Are a Termbase and a Translation Memory For?

A translation memory is a database that stores pairs of previously translated sentences. When a new sentence arrives, it checks whether the memory holds a similar sentence and, if it does, proposes the earlier translation. The aim is not to translate the same or similar sentences twice, which gains both time and consistency.

A termbase, by contrast, is concerned not with sentences but with individual concepts. It holds each record as a concept together with its preferred equivalent, its synonyms, its subject field and its definition. The aim is to make sure a concept is called by the same term everywhere. Hann’s dictionary is the classic example of a termbase on paper; it organises concepts by subject code.

The difference lies in scale. The memory works at sentence level, the termbase at word and concept level. Even when a sentence changes completely, the terms inside it may stay the same, and that is exactly why the termbase steps in where the memory misses. The two are like a pair of spectacles ground to different resolutions.

What Do They Store? Terms or Sentences?

The unit of a translation memory is the sentence, or in more technical terms the “segment”. The source sentence and its approved translation are stored as a pair. As the memory grows, it produces more and more suggestions for repeated and similar sentences. Those suggestions, however, rest on sentence similarity alone; they carry no information at the level of meaning or concept.

Dimension	Translation memory	Termbase
Unit	Sentence (segment)	Concept/term
Aim	Reusing the sentence	Standardising the term
Formation	Accumulates over time	Built deliberately
Information	Sentence similarity	Concept and context

The unit of a termbase is the concept. Each record carries a concept together with its equivalents in the various languages, its status and its subject field. This is why a termbase is useful even when no sentence repeats at all: the sentence may be new, but the terms in it are familiar. The memory stores history, the termbase stores decisions.

This distinction also explains why the two tools are worth different amounts in different types of text. In a highly repetitive user manual a translation memory pays off handsomely, because the same warnings and the same steps keep coming round. In a marketing or legal text where every sentence is unique, by contrast, the memory contributes little. A termbase produces the same value in both types, because concepts recur even when the sentences are original.

How Do They Work? The Logic of Matching

A translation memory works on fuzzy matching. It compares the new sentence with the sentences in the memory and calculates a similarity percentage. On an exact match the earlier translation comes straight through; on a partial match the translator adapts the suggestion before using it. Once similarity falls below a given threshold, the memory stays silent.

A termbase, on the other hand, matches at word level. When a term held in the termbase occurs in the sentence being translated, the equivalent the termbase proposes is shown to the translator. This takes no interest in the sentence as a whole; it simply flags the recognised terms inside it. The translator therefore finds the most critical choices ready made.

The two mechanisms are powerful when they run together. The memory brings in the skeleton of the sentence, while the termbase guarantees that the terms inside it are correct. If an old sentence proposed by the memory contains a wrong term, the termbase catches it. In this way each tool closes the other’s blind spot.

How Do They Complete Each Other?

A termbase and a translation memory are partners, not rivals. When the termbase enforces consistent terms, the sentences that enter the memory are consistent too, and that makes the memory’s later matches cleaner. Conversely, when the memory stores approved sentences, the way terms are used in those sentences can feed the termbase.

This complementarity shows up most clearly on large, long-running projects. Over a product documentation cycle that lasts years, the memory catches the repetitions and brings the cost down, while the termbase keeps the product’s terms identical from release to release. One carries productivity, the other consistency; together they build quality.

Consider a concrete example: the manual for a new version of a device is being updated. The paragraphs that have changed are new, the ones that have not are old; the memory pulls the unchanged sentences up at once and stops them being translated again. In the newly written paragraphs there is as yet no counterpart in the memory, but the terms that occur in those paragraphs are recorded in the termbase. The new text is therefore written using the same terms as the old. The two tools divide the work between two different parts of the same document.

Using only one of them is half a solution. A team working with the memory alone drifts in its choice of terms in new sentences. A team working with the termbase alone cannot make use of sentence repetition. The material in our knowledge centre illustrates how to set the two up together. The most productive set-up combines them in a single workflow.

Which One Is Set Up First?

A practical question arises: which should an organisation starting from scratch set up first? A translation memory accumulates over time, as work is done; it is hard to have a full memory from day one. A small preferred term list, on the other hand, produces value even on the first project and can be built straight away.

In most cases, then, it is sounder to start with the termbase. A few dozen critical terms and their preferred equivalents raise consistency immediately. The memory, meanwhile, grows of its own accord in the background with every completed project. A termbase is a deliberate investment, a memory an asset that accrues.

The two can still be run in parallel. While the termbase is being built, a memory can also be compiled from past translations. What matters is that, whichever one you begin with, the other is not neglected. A one-legged set-up sacrifices half the return translation technology offers before it has even started. The order of set-up is a question of priority; leaving one of the two out permanently is a shortcoming.

Common Misconceptions

The commonest misconception is to take the translation memory for a guarantee of terminological consistency. The memory only helps with sentences that have already been translated in the same way; it does not secure the choice of term in a new sentence. An organisation that wants consistency has to lean on the termbase, not on the memory.

The second misconception is to see the termbase as an unnecessary burden. The thought that “the memory remembers everything anyway” collapses in texts with low sentence repetition. In marketing, legal or creative content sentences rarely repeat; there the memory’s contribution is limited and the termbase’s is decisive.

The third misconception is to think the two are the same thing, set one up and skip the other. As we have seen, they solve different problems. You do not have to choose between them; the real loss is ignoring the productivity the two would produce together. The right question is not “which one?” but “how do I set the two up together?”.

A fourth and subtler misconception is to trust the memory without ever checking it. The memory stores old translations, but if one of those old translations contains a past error, the memory will faithfully propose it again. Here the termbase acts as a corrective: if a term in the sentence the memory brings up does not match the termbase, a warning is raised. Blind faith in the memory is the easiest way to carry past errors into the future; the termbase balances that faith with checking.

The Türkiye context

Most organisations in Türkiye that use translation technology know the translation memory but neglect the termbase. Yet the structure of Turkish makes a termbase all the more necessary. Because the language is agglutinative, the same sentence can look different to the memory through small inflectional variations, and this lowers the fuzzy match rate and reduces what the memory contributes.

In this setting the termbase fills the place where the memory stays weak. Even when sentence matching is low, the preferred terms in the termbase carry the consistency. The abundance of synonyms in Turkish also makes the synonym management of a termbase indispensable. At our office in Ankara Kızılay we recommend, on corporate projects, setting the two up together and keeping the termbase permanently open to the translator. The terminological consistency the memory cannot provide is then secured by the termbase.

What does this mean for you?

If you are an organisation investing in translation technology, knowing the difference between the two tools lets you set the right expectations. Expecting from the termbase what the memory cannot solve, and from the memory what the termbase cannot solve, is wasted effort. To get the best result:

1. If consistency is your priority, build a short termbase first.
2. If you have highly repetitive documents, bring a translation memory into play from the outset as well.
3. Use the memory and the termbase together in the same workflow; do not favour one over the other.
4. Keep the termbase open to your translators and your tools.

On corporate projects we set up the termbase and the translation memory together and run the two in step; our translation services cover this set-up. To have us assess your existing infrastructure, you can reach us through the quote form.

Frequently asked questions

Are a termbase and a translation memory the same thing?

No. A translation memory stores previously translated sentences and proposes them for similar sentences. A termbase manages individual concepts and their preferred equivalents. Their functions differ, but they are complementary.

Which should be set up first?

A small preferred term list produces value even before a translation memory does. A memory accumulates over time, whereas a termbase is built deliberately from the start. It is generally sounder to begin with the termbase.

Does a translation memory ensure terminological consistency?

Not on its own. The memory only helps with sentences that have already been translated in the same way; it does not secure the choice of term in new sentences. Consistency calls for a termbase.

Is it essential to use both together?

Not essential, but productivity multiplies when they are used together. The termbase enforces the correct terms and the translation memory reuses the sentences that contain them. Each improves the other’s output; used together, both speed and consistency rise at the same time.

References

• Hann, M. (1992). The Key to Technical Translation, Vol. 2. John Benjamins.
• Bowker, L. (2002). Computer-Aided Translation Technology. University of Ottawa Press.
• ISO 30042:2019. Management of terminology resources – TermBase eXchange (TBX).