Tips & Tricks

The quality of a machine translation is dependent not only on the performance of the translation system, but also on the suitability of the source texts. The latter can be influenced for texts that are destined for translation from the very outset. In the following excerpts, you will find a few tips on how to optimize your texts for translation.

• Avoid run-on sentences.
• Avoid nested sentences.
• Avoid long attributes.
• Simple sentence structure.
• Avoid padding text.
• Use the correct apostrophe symbol.

Avoid run-on sentences.

Short sentences are easier to understand than long sentences. This applies to people as well as to machine translation systems. Of course, the system does not "understand" the content of a sentence, but it is a lot easier for a system to analyze clear, short sentences. This reduces the time needed for translation. And the error quota is also reduced. Long sentences can always be changed into several, shorter sentences. Here, attention needs to be paid to clear sentence connections and references. If this makes the text that little bit longer as a whole, it will cause fewer problems than a text that can only be understood after it has been read twice.

not:	The purpose of the report is to inform you of our decisions for the new development model, which concentrates on the structural and formal aspects of the software development, as well as discusses our new approaches to quality improvement, which consist of discussion cycles, quality assurance and reviews.
better:	The purpose of the report is to inform you of our decisions for the new development model. It concentrates on the structural and formal aspects of the software development. Furthermore, it discusses our new approaches to quality improvement, which consist of discussion cycles, quality assurance and reviews.

Avoid nested sentences.

A lot of thought is usually required to untangle nested sentences. For this reason, they should be avoided in the same manner as run-on sentences. Sentences with several relative clauses should be separated at all costs. Verbs should be brought forward to the beginning of the sentence and there should only be one subordinate clause.

not:	Technical reports, which are documents that describe the process, progress, and results of technical research and also include recommendations and conclusions of the research, are today a major source of scientific and technical information.
better:	Technical reports are today a major source of scientific and technical information. They are documents that describe the process, progress, and results of technical research. They also include recommendations and conclusions of the research.

Avoid long attributes.

Attributes should qualify a noun, an adjective or an adverb. They should not eclipse it. If an attribute is too long, it is unnecessarily difficult for a system to perform syntax analysis. This can result in translation errors. Here also, overlong constructions should be avoided in favor of subordinate clauses or several sentences.

not:	The program GUI, being a movable and scalable window, designed to be a round small shape, when in the middle of the screen, and becoming an oval with straight borders at the edges, when placed at the top or the bottom of the screen, has a user-friendly and modern design.
better:	The program GUI, being a movable and scalable window, has a user-friendly and modern design. It is designed to be a round small shape, when in the middle of the screen, and becomes an oval with straight borders at the edges, when placed at the top or the bottom of the screen.

Simple sentence structure.

Subject, predicate, object: this is the word order in a simple sentence. Of course, texts should not sound monotonous or sterile because this sequence is the only one being used If the object of the sentence needs to be emphasized, it can be placed at the beginning of the sentence. But this should be used selectively, as the machine translation system cannot accommodate it. Sometimes, the meaning can only be recognized by identifying the subject and object of a sentence. However, in contrast to people, the translation system can only apply syntactic and morphologic rules and algorithms. This can be seen in the following extreme example.

not:	Responsible for publishing the reports in 2009, as before, was our house publisher.
better:	Our house publisher was responsible for publishing the reports in 2009, as before.

Avoid padding text.

Padding text contributes nothing to the meaning of a sentence. On the contrary, they often make a simple sentence wishy-washy. The machine translation system tries to assign them a syntactic function which they do not actually have. This can easily lead to translation errors.

not:	The five candidates have now all passed, so to speak, their entry test.
better:	The five candidates have now all passed their entry test.

Use the correct apostrophe symbol

I'm not going to the cinema tonight (use the symbol below the @ symbol!)

Please do not use the following accents:
I `m not going
or
I´m not going.