Cole and Hermon 2008: VP raising in a VOS language #

Toba Batak is verb-initial: the unmarked clause is V-O-S, with indirect objects and adverbials after the subject; direct objects and passive agents cannot be extracted while subjects, indirect objects and adverbials can; and the active subject binds a reflexive object, the passive agent a reflexive passive subject and, on the authors' own fieldwork, the passive subject a reflexive passive agent, while the active object cannot bind a reflexive subject. The paper derives all three from one derivation. The verb adjoins to Voice, the indirect object and then the subject raise out of VoiceP to specifiers of functional heads, and the remnant VoiceP raises over them, so word order is the order of extraction.

Whatever is still inside VoiceP when it raises is frozen there, so the object and the passive agent, which never leave, cannot be extracted. Binding requires c-command at some stage of the derivation, so the passive agent binds from the base and the passive subject after it raises past the agent, while the active object never c-commands the subject. Schachter and Sugamoto's Semantic Hierarchy Condition, actor over patient over dative, licenses the passive agent but cannot separate the passive subject as antecedent, acceptable if less usual (Table 1's Type B), from the active object as antecedent, ill-formed for every speaker (Type C); surface c-command fits neither VOS passive and reconstruction to the last stage before VoiceP raises only one of them.

Surface SVO clauses raise the subject once more, past the fronted VoiceP, so they freeze the object as VOS clauses do, against the hypothesis that VoiceP stays in place. The sides at which Merge attaches the material inside vP change neither the derived objects nor the surface order, which the paper takes as grounds to propose that Merge is unordered and order a by-product of Move. English, whose passive agent is an adjunct below the patient, lets the patient bind it but never the converse.

The derivation is a Derivation on the syntactic-object carrier, freezing and binding at a stage are the substrate's Derivation.ObeysFreezing, CCommandsAt and BindsAtSomeStage, and the paper's examples are the rows: vos_remnant says that the raised VoiceP is a remnant, vos_hypothesis_only and derivational_only single out, among the candidate analyses of word order and of binding, the paper's own, Licenses.derivational shows the paper's binding condition the weakest of the configurational ones, table1 derives its three grades, stageAt_sides and surface_sides are the §6 claim over every attachment choice, and english_passive the §7 contrast.

Implementation notes #

The clause is a token of the paper's transitive frame on the carrier; the substrate's clause types serve other theories (Landau's inflection–complementizer pair, valency frames). Its slots are the two positions of vP, ArgPosition, extended by the ditransitive goal, and the paper's roles are read off them.
Voice and v are separate heads (fn. 20). The carrier has no head adjunction: the verb and then Voice re-Merge at the VoiceP root, standing in for the two raisings of (51)–(52) that the paper draws as one right adjunction (53), so the raised VoiceP is {Voice, {V, {t, vP}}} rather than a complex head over vP; nothing about the DPs depends on the difference. The functional heads F, whether T or Asp (fn. 22), are Cat.T; the paper's unfilled XP in Spec,VoiceP is omitted.
The replay externalizes leaf Merge only, so a goal PP enters as one leaf, as the paper fronts it. Wh-fronting is C's Merge and the raising of the argument to its specifier, and Freezing reads the movers, so the remnant, carrying the evacuees' traces, freezes only what stayed in it; fn. 18's caveat on extraction from moved complements is the paper's.
Binding at a stage is decided on the replay, which forgets to the derivation's own stage; a stage whose replay fails c-commands nothing, and the surface-order theorems witness that every derivation here replays.
The rows record their configuration in paper features; the theorems decode it into a schematic clause and compare the derivation's prediction with the row's judgment.

TODO #

Adverbials ((15)–(16), (47)–(48)), which fn. 25 evacuates like the goal, the NP- and PP-internal antecedents of §3.6 ((30)–(34)) and the object–goal data of §3.4 ((18)–(20), (22)–(23)) are not modelled, since a PP is a leaf.
Fn. 28 offers two readings of the Type A over Type B preference, a preference for reconstruction or the Semantic Hierarchy as a separate constraint; only the second is formalized.
The paper's conditional remark that its English passive excludes Collins's smuggling derivation is prose; a smuggling derivation on the carrier belongs with Collins 2005.

Cole and Hermon 2008: VP raising in a VOS language #

Implementation notes #

TODO #

References #

The clause and its derivations (§4.1–§4.2) #

Word order (§3.1, §3.3, §4.1, §4.2, §5) #

Every clause goes through a VOS stage (§5) #

The rows #

Extraction (§3.2, §4, §5) #

Binding (§3.4–§3.7, §4.3) #

Merge is unordered (§6) #

English passives (§7) #