How (not) to derive a *ABA: The case of Blansitt’s generalisation

In this paper, I provide an account for the so-called Blansitt’s generalisation (Blansitt 1988). The generalisation says that in the linear sequence DATIVE—ALLATIVE—LOCATIVE, only adjacent functions may be marked the same. In previous work (Bobaljik 2012; Starke 2009; Caha 2009), analogous *ABA patterns have been encoded by the so-called feature cumulation. Feature cumulation means that the amount of features characteristic for individual categories monotonically grows in the order given in any such sequence. However, Blansitt observes that in the case of datives, allatives and locatives, the allative (which is in the middle) tends to be composed of the dative and the locative, so the account based on cumulation does not work. The present paper thus argues for a different representation of the underlying categories, namely as containing (abstractly) the features A, AB and B respectively (following in part Bobaljik & Sauerland 2017). I refer to this as the “overlapping” decomposition. When such a decomposition is combined with the Superset Principle (Starke 2009), it yields both the *ABA restriction and the observed syncretism and containment patterns. I further argue that this is a non-trivial result, since traditional underspecification approaches face significant challenges in deriving the *ABA pattern on the basis of the overlapping decomposition. This article is part of the Special Collection: *ABA