Replacive grammatical tone in the Dogon languages
The Dogon languages of Mali display a complex system of grammatical tone, wherein a word’s lexical tone is completely replaced by a grammatically conditioned overlay in certain morphosyntactic contexts. For example, the sequence Noun Adj may surface as NounL Adj (the noun takes a {L} overlay), while the sequence Possessor Noun may surface as Possessor NounHL (the noun takes a {HL} overlay). The question arises as to what happens when two or more overlays come into conflict, as in the sequence Possessor Noun Adj. The answer to this question depends on the language. In Tommo So, the adjective’s overlay takes precedence (Poss NL Adj), while in Nanga, the possessor’s overlay is the stronger (Poss NHL Adj). In this talk, I present original data from a sample of nine Dogon languages showing that the system of replacive tone depends on both syntax and morphophonology. Specifically, I argue that tonal overlays are implemented in the grammar by constraints whose scope is dependent upon syntactic structure in the form of c-command. I work within the framework of maximum entropy grammar (Goldwater and Johnson 2003), which uses weighted constraints rather than ranked constraints of strict Optimality Theory (Prince and Smolensky 1993), to account for languages with variable output forms. For example, a constraint Adj,L would assign a violation to any word c- commanded by the adjective that does not take a {L} overlay. If this constraint has a higher weight than the possessor’s constraint, Poss,HL, then the {L} overlay will take precedence; if the possessor’s constraint is more heavily weighted, the {HL} overlay will take precedence. I show that the dazzling complexity found in the Dogon language family arises naturally from re-weighting a common set of constraints, lending strong support to this constraint-based model.