Indeed I think we should have the translation strings break per paragraph instead of text parts. I think a solution would be to use a gettext style and use the Genshi Translator.extract on the stream (instead of the ODT XML).
> We verify that all translations use this terms: http://www.europeanpaymentscouncil.eu/index.cfm/other/core-sdd-mandate-translations/
> Maybe the report must not be translated as others but have a copy for each country-language.
Seems the only reasonable way. A feature like calling <report_name>_<xx_XX>.odt (where xx_XX would be the localization of the recipient/destination party) would be a useful feature for all kinds of legal texts.