From a linguistic perspective, no language is more efficient than another language. Switching to an Asian language like Chinese would not necessarily be a better representation for the neural network than English. Mandarin Chinese is a very analytical language with a low inflectional morphology, but it is no less complex. For example, it has a large number of modal particles that have no equivalent in English.

In linguistics, there are also attempts to convert languages into other forms of representation. The natural semantic metalanguage (NSM), for example, reduces words to a set of semantic primitives.

I am a bit more skeptical from what I have seen both in linguistics and in NLP.