All communication relies on assumptions, for example about the meanings of words. If someone says: „Pass the salt, please,” then without assumptions, endless clarifications would be needed about the meaning of „pass” and „salt”, about the language used and whether the request is a joke or carries some hidden meaning. If the requester clarifies: „I did not mean it as a joke, I really want you to give me the salt,” then clarifications about the clarification would result. Again, the clarification may be a joke (we shouldn’t assume it is not) or use words like „not” in an ironic way with meaning the opposite of the usual (we shouldn’t assume the usual meaning). There would be an endless cycle of clarifying the clarifications of… of clarifications.
Different languages use different phrasing to express the same level of politeness, formality or familiarity. Directly translating the words from one language to another may result in a sentence at a different (usually unintended) level. This creates the impression that the speakers of a language (usually coinciding with people from a particular country and culture) are all polite or all rude. They simply translate their thought to another language, where the phrasing is above or below the intended level of politeness. This is one way that stereotypes about a nationality’s modesty, assertiveness or formality are created.
As an Estonian in the UK, I was considered rude, because I answered “yes” or “no” instead of “yes, please” or “no, thanks.” In Estonian at the time, “no, thank you” would have been formal or ironic (mockingly formal) and I just translated my ordinary reply to English.
The level of formality in a language changes over time and with social class. A good example is TV series about 19th century Britain, where people say things like “you are too kind, Sir”. In modern times, this would sound strange.
If a text has already been translated to a couple of languages with high quality, then it may be possible to improve the quality of machine translation to another language by translating separately from each original language and averaging in some sense. I do not know whether a program currently exists that is able to take into account multiple starting languages – Google Translate and other online automatic translation services I have seen only use one. Several different translations should contain more information than one, so by comparing them, some errors may be eliminated. At least inconsistencies can be discovered by computer and then checked by a human, saving labour.