Translation Issues 3-99 (September 1999) by Harald Stücker

The translation market is growing with tremendous speed. Pressure comes from various angles: volume, time, quality and price. Hence the challenge can be stated thus: Translate more better and in less time at a lower cost! There is no way this can be done without the use of translation tools.

Following standard terminology, I will distinguish pure Machine Translation (MT) from Machine Aided Translation (MAT). We may subdivide the latter further into Human Aided Machine Translation (HAMT) and Machine Aided Human Translation (MAHT):

• HAMT: The core job is done by an MT engine. The human translator helps with preparing terminology (coding) and the text (pre-editing) and repairing (post-editing) the text.
• MAHT: The core job is done by the human translator. A machine may help with terminology management and translation memory (TM).

 

Internationalisation

Translation as an afterthought is still widespread. As a multilingual documentation consultant, however, we keep emphasising the unity and interdependency of the whole process, and drawing attention towards its beginning. This is what we may call the internationalisation aspect of documentation.

By internationalisation we understand the design of a (software) product with localisation at a later stage in mind. If language resources are not strictly kept separate from code, localisation will be difficult, error-prone and - with big and complex products - may eventually turn out to be impossible. In general, localisation as an afterthought will be very expensive in terms not only of cost, but also of time and quality.

The same applies to documentation in general. Because internationalisation is an issue wherever products contain language or location-specific information, it is always an issue for documentation. This is especially true if MAT is used.

 

(MA)HT or (HA)MT

Before you start authoring, you should decide whether and how you will translate the text. The distinction put forward by Jeff Allen (TC-Forum 1-99, p. 5) between texts destined or chosen for translation is crucial here. The degree of rigour you have to employ on authoring depends on how much you want the machine to do.

 

Using (MA)HT

If you know that a core use of MT is excluded, you may allow for a lesser degree of rigour in authoring. Here are some points to keep in mind concerning Translation Memory (TM) systems:

• Create a monolingual terminology database before authoring. This is especially important where the documentation is written by more than one author. The use is threefold:

1. your source documentation will be more consistent,
2. multilingual terminology will be more easily created, and
3. because of (1) and (2), your translated documentation will be more consistent.

• Loose talk about matching can be misleading. Strictly speaking, there is no such thing as a 100 % match. Translation is concerned with content/context, but "match" is a purely formal and statistical category. You will always have to double check.
• It is widely held to be true that not all texts are suitable for TM processing. This is not obvious. Given effective content management, even less repetitive, more creative types of text (like marketing or collateral material) may profit from a TM connected to the product family.
• Layout changes in a new version may lead to poor matches or much post-editing effort to reformat tags. Use a template system and allow layout changes only on template level. Basically, the separation of form and content is what internationalisation is all about.

 

Using (HA)MT

If the machine will take over the core part of the work, the text has to be authored more rigorously as a destined-for-translation-text:

• You should use some kind of Controlled Language (CL).
• You have to perform a different kind of terminology work. You not only have to translate the original monolingual terminology, but also have to code syntactic and semantic information.
• A properly CL-authored text may need only very little pre-editing. The main task will be to protect text that is not to be translated. For every text not properly authored or only chosen for translation, the pre-editing stage involves a re-authoring.
• Post-editing is a very different task from translating. Especially you cannot apply the principle of charity (i.e. "if you don't understand immediately, presume the author intended to make sense and try again!") towards a machine output. At worst, given the wrong process, post-editing can be much more time-consuming than translating from scratch.

To summarise: In a multilingual documentation process involving n languages, you will have to pay no less than n times for each mistake committed at authoring stage. Internationalising documentation pays.

"transline Deutschland - Übersetzungsdienst für technische Übersetzung"