Most split alignment systems only have split alignment in the case marking, and not really in the underlying syntax. The split often also either correlates to the TAM of the verb, or to certain grammatical persons. I don't recall seeing any other actual cause of an alignment split, and I would really be interested in hearing about any other triggers for it.
However, that's not what I am writing this post about. This post is about a split in syntactical alignment based on a semantic distinction among verbs. I further find this particular split rather likely to occur.
The particular thing I'd think would cause an intransitive subject to align with an object syntactically is existentialness. Verbs such as 'exist', 'be' (when used existentially), be found, be seen (maybe), be attested, etc all seem to lend themselves well to prefer an ergative syntax.
The most obvious phenomenon to investigate would be gaps, and we can easily imagine a language where
berries exist and I eat
would parse as
berries exist and I eat [them]
however, we can then expect that coordination over many VPs should also yield ergative patterns:
berries existed and I saw and ate and tasted sweetberries existed, and I saw and ate [them] and [they] tasted sweet
This should not be permitted when all verbs are just 'plain vanilla intransitives'
*the berries were tasty and I saw and ate and tasted sweet
*the berries were tasty and *I saw and *ate and *I tasted sweet
Of course, when multiple VPs are chained like this, the later the existential verb appears, the harder parsing correctly will be. For this reason, the language might either forbid existential verbs to be on the right hand of coordinations, especially after more than two or so, or have some kind of 'de-existential' form that is semantically, but not syntactically existential. (This could be achieved by reusing some other thing from the language - maybe force quirky case existential subjects? maybe use usually non-finite verb forms? maybe have the usual existential verbs lack congruence, but forcing congruence on them turns them into 'regular' verbs?
