Abstract
Agreement systems often allow alternatives: This family has / have lost everything. Therefore typology requires a means for generalizing over them. Instances like plural have are frequently termed " semantic agreement " (vs. " syntactic agreement " for singular has), but this notion has proved difficult. The challenge is to encompass the full typological range of alternative agreements. These include the core instances: (i) hybrid nouns like family; and (ii) constructional mismatches, such as conjoined nominal phrases, but also less obvious phenomena: (iii) split hybrids where neither alternative is straightforwardly semantic, both appear related to form, and (iv) examples like Scandinavian " pancake sentences " , which stretch semantic agreement towards pragmatics. These different types are comparable in that (i) the alternatives are realized by the normal agreement forms; and (ii) they are subject to the Agreement Hierarchy. Hence they demand a common treatment. To achieve this, I first unpack the Agreement Hierarchy constraint into the agreement target positions and the directionality implied by " semantic agreement ". I show how the latter arises from mismatches between the agreement information available from different sources. Typically, in the core instances, the information from one source is more evidently semantic than from the second. But in other instances, this is less clear.