Every text is different.
Choose the right route.

Machine translation was long used mainly to obtain a rough understanding of foreign-language content. Globalisation, shorter product cycles and growing text volumes encouraged further applications. Neural methods and newer model generations broaden the possibilities. This does not make them equally suitable for every type of text.

Emails, chats, reviews and other information may be suitable for initial orientation. Misunderstandings must be tolerable and data sharing must be permitted. A short text is not automatically simple; missing context can be crucial when there are only a few words. Binding decisions should not rely on unchecked translations.

Product information, manuals and business correspondence need a quality scope suited to their use. Light post-editing aims at agreed basic usability; full post-editing also addresses the defined linguistic and stylistic requirements. For legal, safety-related or particularly creative text, we assess the necessary specialist and editorial process.

Text needs to be available in a processable form. Scanned PDFs require text recognition and checking; audio requires transcription. Layout, tables, drawings and embedded text may need additional steps. The same applies to multimedia: a translation feature does not automatically handle editing, timing or checks against visuals.

Not every language pair and subject area is equally well represented in a model. We select a representative sample, define review criteria and compare error patterns, editing effort and technical steps. This establishes whether machine pretranslation, another workflow or a combination is suitable. General percentages for time or cost savings do not replace that assessment.