Song (1996): Causatives and causation #

[Son96] classifies causative constructions by what joins the expression of cause to the expression of effect. In the COMPACT type [Vcause] and [Veffect] share one clause, fused (English kill), bound (Turkish öl-dür) or free (French faire lire); in the AND type a clause of cause and a clause of effect are coordinated, in that order (Vata le); in the PURP type the clause of effect is a clause of purpose (Korean -ke) (chapter 2). The traditional typology, lexical, morphological and syntactic, splits the COMPACT type and lumps the other two together (§1.2). The COMPACT type being the diachronic residue of the other two, what distinguishes those is implicativity: the AND type expresses the EVENT and RESULT stages of the cognitive structure of causation GOAL → EVENT → RESULT and entails its effect, the PURP type expresses GOAL and EVENT and does not, except through the inference that people realize the goals they act for (chapter 5). The grammatical relation of the causee follows no hierarchy of its own: a language caps the core NPs of a simplex clause, causative or not, and its causatives keep within the cap (chapter 6).

Form.type computes the type of a form and Form.traditional its traditional class, and on Song's examples neither classification refines the other (not_factorsThrough_type, not_factorsThrough_traditional). A biclausal type entails its effect in every model exactly when it expresses RESULT (forall_meaning_subset_iff), so the AND type is implicative and the PURP type is not (and_meaning_subset, purp_not_implicative), as the example (7) bears out (effectNegated_rows), while COMPACT causatives go either way (compact_implicativity_varies); the success assumption restores implicativity (meaning_subset_of_success). The paradigm causative of [Com89]'s case hierarchy never has more than three core NPs (paradigmCoreNPs_le_three), so a language capped at two refutes the hierarchy for transitive bases (paradigmCoreNPs_two).

Implementation notes #

A form records the structure Song reads off an example: one clause and how [Vcause] is realized, or two clauses, their link and their order. The forms of the examples are row features, and their types are computed.
The traditional class of a free [Vcause] is morphological, after [Com89]'s treatment of the French causative as quasi-morphological, which Song reports (p. 161); Song's own description of the traditional typology (p. 135) does not place it.
The semantic structures are those of the prototypical types. A model interprets the three stages by arbitrary propositions, so that entailing the effect is doing so in every model. The cline of implicativity within the PURP type (p. 136; a survey of sixteen Korean speakers, §4.5) and the implicativity of COMPACT causatives are outside the model.
paradigmCoreNPs counts the indirect object as a core NP, as in a language capped at three (p. 174). That the paradigm case keeps within a cap of three is drawn here rather than stated by Song, who argues the reinterpretation for extended demotion and doubling (§6.5).

TODO #

The diachronic model of causative affixes (chapter 3): four stages from a construction whose [Vcause] is obligatory to a causative affix, and the argument that PURP, overt and meaningful, can take over [Vcause] while AND, often covert and empty, cannot (pp. 80–85).
The measures by which causatives keep within the cap (§§6.4–6.6): omission, demotion to an adjunct, doubling and detransitivization.

Song (1996): Causatives and causation #

Implementation notes #

TODO #

References #

Forms and types (chapter 2) #

Implicativity and the cognitive structure of causation (chapter 5) #

NP density control and the case hierarchy (chapter 6) #