This page answers questions about the use of automatic translation. The point most often misunderstood is this: the fact that automatic output is fluent does not mean that it is correct.
Do you use machine translation?
The scope of use and the confidentiality conditions are discussed at the outset. No client document is uploaded to uncontrolled services. The decision is made in writing.
Why are uncontrolled services not used?
Where the document goes and how it is stored cannot be controlled. This is unacceptable in terms of confidentiality and data protection. The rule has no exceptions.
Can we use it in our own system?
A controlled installation is assessed differently. The conditions are defined together. The decision is made according to the type of document.
Is automatic translation suitable for every job?
It is not. It is not considered suitable for legal and medical texts or for texts that will be published. The decision on scope is made according to risk.
The output looks fluent; why is it inadequate?
Fluency is not an indicator of accuracy. Automatic output can be fluently wrong, and this is the error that is hardest to notice. Review is mandatory.
What does post-editing mean?
It is the correction of automatic output by a human. It is a different job from the revision of a human translation. The scope is defined separately.
Is this method cheaper?
It depends on the type of text. In some texts it brings a gain; in others the correction effort is greater than translating from scratch. A preliminary assessment is made.
In which texts does it work?
It gives better results in technical texts that are repetitive, formulaic and written in controlled language. It is not suitable for creative texts. The distinction is clear.
Can the level of correction be chosen?
It can. Light post-editing done only for comprehensibility and the publication-ready level are different. The level is agreed in writing.
Will the result be the same as a human translation?
It depends on which level is targeted, and this is written down at the outset. If expectations are not aligned, a dispute arises on delivery. The scope is made clear.
Is it the same thing as a translation memory?
It is not. A memory suggests sentences previously translated by a human; automatic translation produces new text. The two sources are flagged separately.
Does automatic output get mixed into our memory?
To prevent this, the records are flagged and kept apart. An unflagged record misleads in later jobs. The distinction is preserved.
Is it used for documents containing personal data?
It is not. For these documents the prohibition is even stricter. The rule is applied.
Do you inform us that it has been used?
We do. The scope and level of use are agreed in writing. A concealed method costs trust.
Does the quality measurement change?
The criterion is the same, but the expected level is determined by the decision on scope. The threshold is defined accordingly. The assessment is read together with the scope.
Is our terminology list used in automatic translation?
In controlled installations the list can be applied and compliance is scanned separately. In uncontrolled services there is no such assurance. The difference is decisive.
Will everything become automated in the future?
The tools are developing, but the dimensions of context, responsibility and confidentiality require human decision. Our approach looks not at the tool but at the risk of the job. The decision is made per document.