On plain texts and ciphers (a thought experiment)

[Emma May Smith]

...
Is there anything in Voynichese that argues against this?
...

It depends on whether or not you consider the word breaks as natural breaks.

If you assume they are, ask yourself the question, "How many natural languages have certain letters that occur only at the beginnings or endings of words and never (or almost never) anywhere else, compounded by the fact that certain letters occur only in the middles of words and never anywhere else?"

Do you mean script or language? There are multiple scripts with different forms for characters depending on where in a word they occur: Arabic, Hebrew (limited), Syriac, Mongolian.

If you mean language, then we're speaking about sounds, and the answer heavily depends on what you believe the words represent. If you think the words are single syllables, then there are definitely many, many languages which only allow certain sounds at the beginning or ends of syllables, and there is essentially no problem.

If the words are multiple syllables then the answer is more complicated but still ultimately resolvable. Some characters (I'm thinking particularly of [i] and [n]) could represent vowel quality and only occur in stressed positions. If stress is mostly fixed on the final syllable then these characters will mostly occur at or near the end of a word. The character [q] may represent a sound such as /h/ which is only realized in a very limited environment and deleted elsewhere.

You also have to ask yourself whether any of the characters are null characters. Unless the subtle differences in letter shapes are representative of different letters, the alphabet is very abbreviated, about 17 to 20 common characters depending on how you evaluate them. If you then assign some of them as nulls, the alphabet becomes even smaller. It would be difficult to represent a natural language with so few sounds, unless it were an abjad or some similar system.

You also have to account for the very high level of repetition... not just when the same words are next to each other (which might happen in poetry or lists) but the high level of similarity of letter order in words that occur frequently.

I don't think null characters are a problem of natural language theories. Indeed, it is the opposite, with most natural language theories I've read attempting to incorporate the whole text, and cryptological theories using null characters.

But your two points seem to potentially answer each other: too few characters to fully represent all sounds and lots of repetition. This is suggestive that the characters may represent the sound of the language only imperfectly, not drawing all the distinctions that are present in speech. The reader may have had to fill in some details such as the precise quality of a vowel or the voicing of a consonant. This is not unknown in examples from history.