HOW DOES TRANSLATION MEMORY WORK?
From early on, there has been craving for harnessing computers to serve in translation tasks. The first attempts date to the 1950s. Back then the motivation was the Cold War and defence needs. The goal was to develop a machine-translation system that would automatically translate the enemy’s messages to one’s own language.
Afterwards, for a long time the focus has been on developing computer-assisted translation systems instead of automated translation. In these solutions the translator utilises information-technology tools, such as translation memory and term management tools specifically designed to make translation work more efficient. Translation memory software does not produce translations automatically. It rather creates suggestions based on previous translations by searching the texts for equivalences on a clause and sentence level.
The first commercial translation memory applications appeared in the 1980s. Since the 1990s, translation memory software has been used more extensively, e.g. in professional translation associations and other organisations which produce vast amounts of translations, such as the European Union.
Translation memory as additional memory for translator
A translation memory programme parses a text to be translated into clause- or sentence-level units which are called segments. When the translator starts translating the text, the programme offers sections of text for translation one segment at a time. Whenever the translator moves to the next segment after having translated the previous one, the programme saves the text along with the translation in its database. When the same or a similar text segment is encountered again, the programme knows to suggest the translation completed previously.
The extent of the benefit offered by the translation memory depends on the nature of the text. The more internal repetition the text has or the more similar the texts to be translated are, the more useful the memory is. The translation memory suggestions can be very useful when translating a user guide or a product catalogue of a certain product manufacturer. Literary translators, however, do not traditionally use translation memory software.
The use of translation memories is beneficial to both the translator and the commissioner of the translation. The more the translation memory can be exploited, the lower the translation costs. Use of a translation memory tool also ensures that the translated text will be cohesive, and that the same terms and concepts that are used in previous translations are also used in the following translations.
Even translation memory has its limits
A traditional translation memory always requires a previous translation as a basis in order to produce any kind of translation suggestions. When implementing a translation memory programme for a new customer or a project, the memory is always empty at first – creating a memory starts from scratch. During the very first phases, the memory tool does not offer the translator much help. However, the benefits become evident when making progress in the translation work.
Sometimes the translation memory can be created from a previous translation which has not originally been translated using a translation memory tool. If an original source text and its translation are both available in electronic format, then the segments of the original text and its translation can be aligned manually. However, the costs should be taken into consideration here, as the alignment work is manual and time-consuming.
In addition, the translation memory needs attention. In principle, the larger the translation memory, the more beneficial it can be. But the management of large memories does require great discipline.
If the same memory is used for translating texts of varying subject matter, then it can occur that the same term receives a different translation depending on the context. When the memory ends up containing different translations, all of which are correct depending on the context, the risk of misunderstanding is always great. Therefore, one should always save information on the correct context along with the translation. Possible inaccurate translations should also be rooted out of the translation memory as early on as possible. Otherwise, they might end up getting repeated in texts.
Next step for translation technology: from memory to intelligence
The restriction of traditional translation-memory programmes is the fact that they only deal with language on a very superficial level. They do not understand anything about the structure of a language and are not able to apply any grammatical information from previous translations. This means that the new text must correspond with the previous text quite accurately in order for the translation-memory software to suggest any translations.
There has been an aim to respond to these challenges by developing new types of translation applications where translation intelligence will be utilised.
A machine translation programme based on translation intelligence has the ability to learn: it is able to apply the information accumulated in the translation database and make deductions on the equivalences between source and target language structures. And where the translation memory is empty at first, the translation intelligence applications usually make use of a prepared general database. They process fairly short units of four or five words, so finding similarities is more likely than when comparing entire clauses or sentences. Shorter units are also less dependent on text type than longer ones. Without exception, a machine translation programme is always able to offer some type of a translation suggestion.
Even if more traditional translation memory applications still belong to the main tools used by professional translators, there lies plenty of potential in the automation of translations. Read more on this in our previous blog article A machine as a translator – what is it suitable for?