I've been closely following the work stemming from St. John and McClelland's Sentence Gestalt Model, which uses a connectist model to extract semantic information about events from sentences without first specifying the syntactic structure.
I've looked at the limitations of the model (see here for a high level overview), but nothing about them looks to be insoluble without the inclusion of syntax. This intuition has been supported by the followups done by Rohde at Stanford and Byrant and Miikkulaine at UT Austin.
So then, what's the point of syntax within the context of understanding events? Is it only epiphenomenal here or have I misunderstood the gravity of the objections to syntax free models of semantics? Does there currently exist a model of event semantics that includes syntax and produces results not reproducible by the models cited here?