Universals constrain change; change results in typological generalizations

case features are assigned configurationally to the Th-roles in a predicate’s Semantic Form. They cannot be idiosyncratic lexical properties of specific Th-roles, and their assignment cannot be specific to particular predicates. This means that the abstract Case feature [+HR] always defines the highest syntactically visible Th-role of a predicate, its “subject”. Given the previously introduced assumption that Th-hierarchy reflects semantic depth, it follows that Th-role reversals, such as a verb “kill” with victim as subject and killer as object, are excluded in principle from any language.8 Another corollary is that “quirky case” cannot be a lexical association of abstract case, but the idiosyncrasy must rather be accounted for at the level of morphosyntactic case. In section ?? provide empirical evidence in favor of this idea. At the level of abstract Case, the proposed decomposition into features has two principal advantages. The first advantage of the feature decomposition is that it makes it possible to individuate exactly the class of grammatical relations which play a role in syntactic constraints, such as binding, control, and parallelism in coordination. For example, the feature [+HR] picks out “A” and “S” in any language, irrespective of its case system, and thus universally defines the relation of grammatical subject, or “external argument”. (51) Def: An argument/Th-role pair bearing the feature [+HR] is a grammatical subject. Note that Uniqueness excludes the possibility that a single predicate might have multiple subjects.9 Categories requiring negative features are syntactically accessible in so far as those features are assigned by the appropriate licensers. For example, transitive subjects acquire the feature specification [–LR] by unifying with a bearer of morphosyntactic ergative case. Consequently, transitive subjects (“A” as opposed to “S” and “O”) can be singled out by the feature bundle [+HR,–LR] in ergative languages, and only in them. This point will be developed in section ??. Secondly, the features provide the appropriate representation on which valency-changing operations are defined. I assume that these operations are triggered by verb morphology in the lexicon. In particular: (52) a. Passive: demotes [+HR]. b. Antipassive: demotes [–HR]. c. Causative: adds/promotes [+HR]. d. Applicative: adds/promotes [–HR]. ditransitive transitive intransitive Th-roles: λx λy λz λx λy λx Abstract Case: [+HR] [ ] [+LR] [+HR] [+LR] [ +HR +LR ] ERG/NOM [–LR] [(–LR)] [(+HR)] [–LR] [ ] [ ] (Ngiyambaa) (ERG) (NOM) (NOM) (ERG) (NOM) (NOM) ERG/ACC/NOM [–LR] [–LR] [(+LR)] [–LR] [ ] [ ] (Wangkumara) (ERG) (ACC) (NOM) (ERG) (NOM) (NOM) The system generalizes correctly to four-place predicates (Donohue 2004). 8This means, for example, that inverse constructions are not passives (contra Perlmutter and Rhodes’ 1988 treatment of Ojibwa but in agreement with Dahlstrom 1985/1991). Some published treatments of psych-verbs are also inconsistent with this assumption. See section ?? below for further discussion. 9Again there are partly unresolved empirical issues at stake. There are at least two types of cases where linguists have been tempted to speak of multiple subjects. One is probably to be analyzed as a multiple topic construction, e.g. Korean (refs.); another consists of derived predicates such as causatives, where the embedded subject (the “causee”) sometimes has “subject properties”. These are cases of multiple predicates, for which we provide layered representations, which permit reference to the abstract Case features of basic verbs to be visible in their derivatives, as discussed in section ??.