Evolution of irregular declensions and conjugations from reconstructed proto-languages

Question

In designing an artificial dialect of the Greek Language, most words and inflections have predictable patterns with a well-guessed/documented evolution which show the origin of the word's roots and stems. This makes it fairly easy to go back in time, change the evolution of the sounds or orthography, and use an existing vocabulary from the already existing Greek Language to create an artificial dialect.

For example, I could thus look at the dialectical declension of a word like σύ, and see that in the Genitive, Attic renders it σοῦ, Ionic σεῦ and others σέο, and infer that the root/stem is σε-, and that the contraction rules show that the irregular declension is σε-ο, then insert my own sound changes based off my own set of rules. Likewise I can find out that σφᾶς is a normal contraction of σφέας or σφέανς with an irregular enclitic of σφε.

What I can't seem to do is take a root that is only used in the dual and transfer it to the plural. In the rare First Person dual, the pronouns used are (NA) νώ and (DG) νῷν. I want to make a plural pronoun using that root in the manner of Latin "nos" instead of ἡμεῖς (which possibly reconstructs from ἀσμέες or the likes.) I can conjecture several possibilities based off a pattern, like (NADG) νῷ, νώς, νῷς, νῶν. But these have no real basis in the Greek evolution from its first ancestors to Classical Greek. I can't use σφεῖς or σφώ or σφωέ to trace a pattern without a good reason, because all of them seem to have their patterns traced from the evolution of ὑμεῖς, σφι, and νώ. Where do irregular declensions come from, and how are they most authentically reproduced in a constructed dialect? I am not a professional, so please give me some leeway if I have made errors presenting this question.

Answer 1

I know some fellow conlangers are much more proficient in Greek and its history than me, therefore I want to concentrate on some more general aspects of the question.

First of all, you are the conlang designer, so you always have the license to do what you want, even in a very naturalistic and diachronic framework.

Second, irregularities can enter a language in many different ways. Very frequent function words tend to acquire phonological simplifications like contractions that the majority of words don't show. One can see suppletion, i.e., the merger of different forms of different origins into one paradigm. One can see cross-borrowing between different dialects, adding to irregularities in sound laws.

Third, it is easier to work in a forward direction in time than backward or sideward (from one dialect to another at the same time). The outcome of sound shifts is usually unique (taking all constraints by the environment into account), but the progenitor of a given form often isn't. So just postulate a plausible progenitor and evolve it according to the rules you set for yourself. If you don't like the results, try another one. You can play around with this. And don't forget the first point: You are the conlang designer, you decide.