Abstract: 

Control has long been employed as diagnostic of subjecthood: in an active embedded clause, the controllee should invariably correspond to the agent qua external argument. Yet, the very notion of subjecthood breaks down in Austronesian-type voice systems where one argument is promoted to “pivot” and triggers a particular morphological form on the verb without altering its valency: subject properties are distributed between the agent irrespective of pivothood, and the pivot irrespective of θ-role, though not in a uniform manner (Schachter 1976; Guilfoyle et al. 1992).

The status of Austronesian voice-marking has been subsumed under three syntactic analyses: (i) pivots are established in vP/VoiceP via a process of object shift and often subsequent A’-Agree reflecting structural case, θ-role or extraction site; (ii) voice-marking reflects absolutive case assignment in a (split-)ergative system; (iii) pivot-hood is a information-structural alternation, resembling (hanging or internal) topichood.

We examine this debate on the basis of control structures, where Austronesian languages are divided into two types: type A, in which control targets the pivot, irrespective of whether it is the agent or not (e.g., Malagasy, Acehnese), versus type B, in which control targets the embedded agent, irrespective of voice-marking (e.g., Tagalog, Madurese). The picture gets murkier through the existence of embedded voice restructuring (involving voice matching and default uses of Agent Voice morphology) as well as backwards- and crossed control configurations in many Austronesian languages. 

At the same time, Austronesian pivots differ with respect to definiteness/ specificity restrictions imposed on them as well as A’-related properties such as reconstruction for principle A, C and quantifier variable binding. These properties again form two types of languages: type C, in which pivots reconstruct and require to receive a definite interpretation (e.g., Malagasy, Tagalog), versus type D, in which promotion to pivot feeds binding and no definiteness requirement is imposed on it (e.g., Acehnese). 

Given that types A & B of control do not fully overlap with types C & D of binding, we delve into the properties of pivots on a language-specific basis, and suggest that their A/A’-nature is neither uniform nor absolute, but spreads over a continuum: some pivots are more A-like elements, some are more A’-elements, and some are mixed, each determined by the type of features involved in the derivation of voice-marking. Such differences might stem from a diachronic transition from A’-to-A syntax of voice marking (cf. Chen & Patrianto 2023), each possibly correlating with several aspects of the clausal structure, such as word order and position of the non-pivot agent.

We conclude that the notion of “pivot” might be a structural epiphenomenon and subject to variation, in which case a uniform syntactic analysis of voice marking should be untenable.