Daniel Harbour (MIT)
The Kiowa Case for Feature Insertion
Background. All theories of morphology must provide an inventory of morphological operations. In postsyntactic theories, as elsewhere, there has been debate about what the inventory of operations should include. Trommer 1997 formalises Distributed Morphology (Halle & Marantz 1993) so that there is a single operation only: vocabulary insertion, introduction of phonological material to morphosyntactic structures. If correct, syntax is mapped directly into phonology and there is no autonomous morphology.
Proposal. I argue that this theory is too restrictive. Morphology must do more than merely insert vocabulary items. It must also insert morphosyntactic features. This predicts the possibility of disagreement between morphosyntax and semantics and, I claim, one sees this in Kiowa, an Oklahoman language (Harrington 1928, Watkins 1984).
Data. Kiowa has a system of rich, fusional, prefixal agreement, representing three arguments: agent, goal, and object. I show that the morphological structure of the prefix is isomorphic to the syntactic argument structure in a way easily capturable by postsyntactic theories like Distributed Morphology. I show also that the four agreement subsystems (intranstive, transitive, and two types of ditransitive) display strong morphological and phonological uniformities, inferring that unified analysis should be given for all four.
A difficulty for a unified approach comes from some ditransitive prefixes. Object agreement has many different allomorphs. Three types of allomorphy are, I argue, conditioned by the feature [-singular] on agents and goals. Sometimes, however, [-sg] cannot be present in the output of the syntax and yet the [-sg] allomorph surfaces. For instance, the áu allomorph for [+sg] objects is generally restricted to ditransitives with [-sg] goals: because 'us' is [-sg], 'he-to.us-it' (dáu) should and does have áu, and because 'you.sg' is [+sg], 'I-to.you.sg-it' (gyá) should not and does not have áu. However, áu unexpectedly emerges in, for instance, 'he-to.you.sg-it' (gáu).
These 'exceptional' prefixes are regular in all other ways, however - note, for instance that both prefixes for second singular goals have high tone and begin with g. So, clearly, they should not be regarded as unanalysable wholes. Feature insertion is the only way to maintain a unified analysis. Moreover, features are correctly predicted to be inserted elsewhere in the language, accounting for further allomorphy effects.
Theoretical ramifications. By permitting feature insertion, do we make morphology too unconstrained? On evidence from null subjects, null objects, and reflexives, I argue that minus is the unmarked value of [±sg]. So, my analysis accords with Noyer's 1998 proposal that only unmarked features are insertible. I show that the need for feature insertion is independent of whether one assumes bivalent or privative features, and that bivalent features lead to a more restricted theory of morphology.
The morphology-semantics disagreement that feature insertion permits has important consequences for the theory of the lexicon and syntax. In particular, it argues against the strong Lexicalist approach to word formation (Lieber 1980, Chomsky 1993), where sentential derivations check off features of presyntactically assembled words: this view predicts the non-existence of Kiowa-type disagreement.
The paper, therefore, sheds light on the inventory of morphological operations, the typology of morphosyntactic features, and the morphology-syntax interface.
© INSTITUT DE LINGÜÍSTICA APLICADA - UNIVERSITAT POMPEU FABRA