Telling allomorphy from agreement

Recent work on allomorphy has tried to propose various notions of locality domains in order to constrain the relation between the trigger and the target of allomorphy. However, unless we have a way to clearly distinguish between allomorphy and cases of syntactic agreement, this approach is bound to fail as one can never tell whether a given alternation is due to agreement or non-local allomorphy. The goal of this paper is thus to provide a set of coherent diagnostics to distinguish the two phenomena empirically. In order to do this, I provide three case studies about phenomena previously analyzed as instances of agreement. For each of these cases, I argue that an analysis in terms of allomorphy is empirically more adequate for a number of reasons. Since two of these case studies involve phenomena where the trigger and the target of allomorphy are not part of the same word, the present paper also substantiates the claim that context-sensitive spell-out phenomena are not restricted to words. Building on these case studies, the final section revisits six diagnostics that can be applied to a given alternation to determine whether it is an instance of allomorphy or agreement.