I seem to have a recurring problem of making conlangs with weird rules that don't work in reality.
I've made several like this. Like a language that relied only on 6 case endings to mark ALL grammatical roles (natlangs that do the same thing typically between a dozen or two). Or my latest attempt: a configurational SOV language with direct-inverse alignment, and zero marking.
To give an idea of how it works, consider the two following gloss translations:
PERSON. 1s. SEE.
1s. PERSON. SEE.
In my conlang, these sentences mean the same thing: I see a person (or saw, or will see, there is no tense or aspect marking). How? Simple, the language has an agentivity hierarchy. If there are two nouns, and one is higher on the heirarchy than the other, then its the agent and the second noun is the patient. Pronouns are above common nouns, so the 1s pronoun must be the agent.
So then, how do you say 'a person sees me'? In a natlang there would be an affix to indicate that the noun lower on the heirarchy is actually the agent, but my conlang doesn't have that. You have to rely on topicalization. The noun is fronted as a topic, and its role in the sentence is indicated with an appropriate 3rd person pronoun. For example:
PERSON, 3s. 1s. SEE.
This would literally translate as 'person, he sees me'. Note that if the two arguments are on the same level in the hierarchy (like our two pronouns in the latter example), then word order alone determines role. Normally, the pronouns would be on separate levels of the hierarchy, but if this language had that then there would be no way to indicate the inverse.
I am starting to wonder though if such a language simply can't work in reality. There is a reason why all known confrontational languages are verb-medial. Because if you have all the arguments appear on the same side of the verb, then there's no amount of rules you could have that would absolutely eliminate the possibility of ambiguity. Using a verb-medial order is simply the only convenient way to consistently differentiate subject and direct object, assuming that neither is marked for case.
I'm not sure my conlang will have that problem, but at the same time I'm afraid to put time into exploring it further considering that it may turn out to be yet another conlang that simply can't work.
Does anyone have any thoughts on this? And just so we know, this is an artlang. Its obviously not an auxlang attempt. If you care to know what it is, its simply my speculation of what the first language may have looked like. Its way there's no grammatical markings. I'm assuming that my fictional proto-humans wouldn't be able to grasp something so abstract. The closest thing to a grammatical particle the language has is the word for 'not', though I'm thinking of making it a verb like in Finnish.