Actual natural language is very fluid. Spelling, pronunciation, grammar, all shift around, and for all manner of reasons. But the main one is efficiency. Humans are always subconsciously trying to cram as much information into as little time as possible, while preserving enough intent. This means extra redundant information whittles away, but there has to be enough redundancy for error correction.

With these highly compressed languages, the problem becomes that a slight mispronounciation or misspelling can totally mess with the meaning in ways which lead to confusion. There are just too many phonemic sounds (a given sound difference is phonemic if the distinction matters, eg in English, p and p' (aspiration) is non-phonemic, but in Korean it is)

When you actually analyze languages for their "baud rate" so to speak, languages with more symbol diversity (English) have less phonemes per unit time, while languages with less symbol diversity (Italian) have more phonemes, but the overall entropy per unit time works out pretty much the same. (I can't find my exact source right now, but if you poke around keywords like spoken language entropy rate, there's quite a bunch).

Basically when you have machine code as a language, you pretty much need machine brain to process it.

Edit: here's a start of some sources:

https://seantrott.github.io/information/

Which mostly cites https://www.science.org/doi/10.1126/sciadv.aaw2594