Halliday & Matthiessen (2014: 568n):
Non-final verbal groups may realise features that relate to the transitivity of the clause (and this is always the case with causative constructions). For example, while phase does not constrain the interpretation of the Subject as a particular type of participant, connation does; it implies that the Subject is like a Behaver in addition to whatever other participant roles it serves. Contrast [phase:] she seemed to like him with [conation:] she tried to like him. This is why phased existential clauses present no problem, but conative ones are odd: there seemed to be a person on top of the hill vs. there tried to be a person on top of the hill. As we have seen in Chapter 5, there in existential clauses serves as Subject, but it does not play a participant role; so when a participant role interpretation is ‘imposed’ by a conative verbal group complex, there is a clash in the grammar.