Abstract
Kroeger (1993) proposed that extraction in Tagalog -- which covers both wh-type and raising constructions -- has these properties: (i) what is extracted must be a subject, and (ii) any clause out of which extraction takes place must be the subject of the higher verb. He proposed that the dependency path is simply (SUBJ+). He also argued that Nominative NPs (ang/si NPs) in Tagalog are subjects, and the subject status of a particular argument is also cross-referenced by the voice-marking on the verb.
In this paper I use evidence from subject raising (see e.g., Dell (1981), Kroeger (1993), Nakamura (2000), Sells (2000)) to offer a reevaluation of Kroeger's (i) and (ii). For while Kroeger's analysis has catalyzed much subsequent research, it falters in Tagalog clauses which lack the core clustering properties, such as Recent Perfective (RP) clauses. These are well-known for lacking any Nominative argument, even though they are finite (see Schachter and Otanes (1972)). It is a long-standing observation that any argument of an RP clause can be extracted, even though none would show Nominative case in situ (e.g., McGinn (1988)). In the economy-based accounts of Mclachlan (1996) and Nakamura (1996), in the absence of a Nominative, any argument NP may to move, as all clause-internal NPs are equidistant from the landing site of movement. Additionally, Nakamura (1996) notes that a CP argument of an RP verb can host an extraction site, even though that CP would not be designated as the subject of its clause -- if voice-marking correlates with case-marking and determines grammatical relations.
I argue that the correct analysis of Tagalog extraction must entail (1)a/b, which are generalizations about surface case:
Building on the analysis in Sells (2000), in which the voice marking designates one argument as a null subject, I propose that the function of Nominative is precisely to mark a phrase which heads a dependency, either to a null pronominal in its own clause, or to a null pronominal or gap in an embedded clause. (1)b is then a consequence.
For (1)a, we know that a Nominative NP is licensed in a clause with a null subject. (1)a indicates that an extraction dependency to that null pronominal -- thereby apparently displacing the Nominative -- is preferred over a dependency to some other argument, which would have to be a standard gap-type dependency. In an RP clause where no pronominal subject is determined by the voice-marking, any argument can be extracted, as a gap-type dependency. I propose reducing (1)a to (2):
Under this view, a typological property of Tagalog is not that its dependency paths are SUBJ+ rather than GF+, but that it prefers anaphoric identification (to the null subject, if there is one) over functional identification in such dependencies.