”The fact that all normal children acquire
essentially comparable grammar of great complexity with
remarkable rapidity suggests that human beings are
somehow specially designed to do this.” (Chomsky, 1959)
That is very interesting. After some research, we discovered that American English has 23,800,000,000,000,000 possible phoneme combinations. Of course, we do not use them all.
But what if a machine could? What if machines could use artificial intelligence to generate disposable languages with impossible grammars? Such languages could be sent in the clear and would be hopeless to break, especially if they were only used once, like a one-time-pad, with grammars that are truly impossible for humans to make sense of.
The computers, using artificial intelligence, could learn a disposable language quickly, charging the grammar with randomness from random-number generation, and from associations that humans do not make.
If computers can use artificial intelligence to create disposable languages with impossible grammars, such languages would be a collection nightmare. Such traffic would be different from traffic that is encrypted. It would immediately be subject to analysis, but such an effort would probably yield little or no information–and be expensive, time-consuming, and frustrating to break. An impossible language could be much less redundant, and that would make cryptanalysis much more difficult. Of course, one could encrypt such traffic with standard means such as wrapping it in a symmetric cipher such as AES256, using PGP, or using a one-time-pad. After the standard encryption was broken or bypassed, and a dictionary of the impossible language were assembled, standard dictionary attacks trying to reach the key, which enciphered such traffic, would be pointless. How would you know when you actually reached the plaintext? Even with guessed plaintext it seems that little could be done, especially if the impossible language is disposed of like a one-time-pad.
Moods such as the subjunctive, imperative, and indicative could be indicated according to the impossible grammar by complex, novel means which would be more than just unexpected, strange, or seemingly contradictory. Irony and other shifts in meaning could be indicated by diacritics or tone. One could also have the computers talk in gibberish to each other at intervals, and then have them break into their disposable language, or stick a disposable-language payload into long iterations of gibberish or the impossible grammar used incorrectly.
Distinguishing between the impossible grammar being used correctly or incorrectly would pose extremely problematic issues for an analyst. The crux of the matter will be how errors are determined. An impossible grammar will be more complex than adding or subtracting kinds of voice, tense. mood, or aspect to verbs; more than declining or not declining nouns; more than using African or Asian phonemes, or, for that matter, bursts of PSK-31.
The point is that it will be impossible for a human to learn. If the impossible-grammar language is truly impossible for a human to learn, its traffic cannot be analyzed until someone refers, in a human-understandable language, to what the impossible-grammar language refers to.