Detail #11: Lexical Hierarchies, again

Some isolating language where a word can be used both verbally and nominally probably already does something at least sufficiently like this to accuse me of plagiarism of real-world languages, but here we go anyway. Assume free word order, and rank words as to likelihood to have different roles in a way where verb, object, and subject are the three primary poles of likelihood, and maybe every word also has an orientation (e.g. verb -> object, verb -> subject, object -> subject, object -> verb, etc) and a distance from the centre. For simplicity, the distance from the centre in the examples will be given as two different: O and o, S and s, V and v, capital standing for greater objecthood/subjecthood/verbhood likelihood, and o,s and v being weaker. The orientation will not distinguish distance. Any word's quality will be given as Position→Orientation. In a situation where several nouns form a sentence, say:

bread_? oven_? coal_?

bread might be O → s→ v, oven O → v → S, and coal s → v → o. We notice coal is the least likely to be an object of these three, and thus we  reassign it to subject:

coal_o→s bread_O→s oven_O→v  coal_S [bread_O→s oven_O→v]

the two remaining alternatives are both equally likely to be objects, apparently, but they both prefer different secondary roles. bread as a subject is already covered, so it cannot be that, and is most likely to be the object, in which case oven would be the verb.

coal_S bread_O oven_V

This would parse as something like "coal bakes/heats (the) bread". It is likely certain tuplets of roots would acquire idiomatic uses that may violate the normal hierarchy-interpretation, and it is also likely certain tuplets of words would develop greater granularity than the average system, due to how often collocations occur in actual speech.

Exceptional "poles" can be considered to exist, but most instrumentals etc are marked by adpositions. A surplus of very similar arguments - such as several humans - can be considered to share the same role, although explicit such marking can be achieved using a conjunction.

I find it likely this violates some universals on account of lack of learnability? Could be an interesting thing to ponder.

EDIT: Rereading this post about two and a half years later, I find that this post never got around to explaining the "shape" of the, well, the hierarchy involved. It's essentially a circle! On this circle reside three points, that are equidistant along the circumference of it. In it, ellipses with a starting point and directions exist:

as you can tell, I am no graphical artist.

Now, in this graph the ellipses (with a direction) depict the probabilities for three words:

you have black corresponding to VoS, i.e. strong likelihood of being a verb relative to other words, followed by low likelihood of being an object relative to other nouns, followed by strong likelihood of being a subject. the red circle starts at S, goes to O and then passes v in a distance. The blueish circle starts at O, and passes s and v at a distance.