Perhaps these birds "parse" compositions in order of most recently received (e.g. as something like a LIFO), or by some other rule that is nevertheless (compositionally) significant.
They conclude that because synthesized cry YX gets a different (null) response than natural cry XY, the compound cry must conform to a compositional syntax. This makes no sense. For one thing, the compound cry carries zero additional meaning beyond that of its components -- as if hearing "the man ate the tiger" were no different from hearing "EAT TIGER MAN". But more importantly, the point of compositional syntax is that components may appear anywhere[1] within the structure of the phrase -- we can use "tiger" as the object of one sentence and the subject of another, or inside a clause embedded in a complex sentence, because the syntax will tell us which is which. All these people have demonstrated is that either (1) there is an ordering constraint on delivering two messages at once; or (2) these birds can tell the difference between recorded calls and synthesized calls.
Compare the English auxiliary verb system, which is non-compositional with ordering constraints. A verb may be modified by any or all of the following: first, a modal ("can try"), second, perfective have ("has tried"), third, progressive be ("were trying"), fourth, passive be or get ("got fired"; "was eaten"). While there are several different options in the first slot, and two (plus absence) in the fourth, it is impossible to substitute a meaningful phrase in for any of these.
[1] Obviously not anywhere, but inside many different component phrases.