How far is Natural semantic metalanguage really natural?

The "natural" part of NSM comes from the fact that all of the items in the vocabulary are claimed to be found in all lexicalised in all languages and thus "universal". This is supposed to establish their status as 'semantic primes' - items that are further inexplainable and from which other meanings can be built.

This in itself is dubious because concepts such as "I, you" may appear as lexical items in all languages but they often do not describe the same oppositions as in English. Also, some of the verbs like 'do' and 'move' do not have the same level of generality across languages - so definitions using them in English tend to be a lot easier to deal with than those formulated in other languages. However, Goddard and Wierzbicka both claim that the primes also have universal 'combinatorial properties' but I'm not convinced.

Of course, there is nothing at all natural about the actual descriptions composed using these semantic primes. Just look at the example cited in Wikipedia which tries to explain the concept of Someone X killed someone Y as:

someone X did something to someone else Y because of this, something happened to Y at the same time because of this, something happened to Y's body because of this, after this Y was not living anymore

This is not natural but is intended to be universal. However, it doesn't take into account the differences in tense/aspect encoding, grammar of causality, etc. So in fact, if those descriptions had to be translated into another language, the translator would still have to interpret what was being said, rather than just mapping primes from one language onto the other. Again, the NSM proponents claim that this is not an issue but it strikes me as quite significant (at least from relatively casual reading of the NSM literature).