Comrie (1989): Language Universals and Linguistic Typology #

Three of the book's chapters make claims sharp enough to state. Case marking (chapter 6) serves first to discriminate the two arguments of a transitive clause: of the five ways the primitives S, A and P can share cases, only the nominative–accusative and ergative–absolutive systems keep A and P apart with two cases, which is why those two are the widespread ones, the tripartite system spending a case on a distinction it does not need and the neutral and the A/P-against-S systems failing to make the one that matters. The same function explains differential marking: the agent of a transitive clause is typically more animate and more definite than its patient, so marking is spent on the departures, a P high on the hierarchies or an A low on them, and each language's marking is monotone in prominence, Russian and Polish by animacy, Turkish and Persian by definiteness, Hindi and Spanish by both, Dyirbal marking its As downward and its Ps upward from the same cut between first and second person pronouns and the rest of the animacy hierarchy.

In causatives (chapter 8) the causee takes the highest position on the hierarchy subject > direct object > indirect object > oblique that the base verb's arguments leave free, as Turkish shows for intransitive, transitive and ditransitive bases; languages may in addition double a filled position, Sanskrit a direct object and Turkish an indirect object, or demote further than required, French by par, and the analysis of the oblique causee as a passive fails for Turkish, where passive applies to every transitive verb but the oblique causee only to ditransitive bases, and for Hungarian, which has an oblique causee and no passive expressing its agent in that case. Where a language offers two cases for the causee, the one lower on the hierarchy encodes the causee that retains more control, Hungarian instrumental over accusative, Japanese dative over accusative, Kannada instrumental over dative; and within a language the more compact a causative's form, the more direct the causation it expresses. Subject (chapter 5) is a prototype, the intersection of agent and topic, so that constructions tied to agents, such as imperative addressee deletion, group S with A even in Dyirbal, constructions tied to affected participants, such as the Nivkh resultative, group S with P, both as tendencies, and constructions tied to neither, such as coordination, vary: S with A in English, S with P in Dyirbal, neither in Chukchi.

The case systems are the equivalence relations on the three primitives, caseSystem_of_equivalence that the five are all of them and discriminates_two_case_iff that two of them are the economical discriminating ones; the book's examples are the rows, over which marking_monotone is the information-flow prediction, causee_rows and passive_analysis_fails the hierarchy against its rival, control_rows the control hierarchy, compactness_rows the directness correlation through the substrate's monotonicity, and bias_rows, coordination_rows and mixed_rows the prototype's predictions.

Implementation notes #

A marking row records, for each argument the book discusses, the scales the book invokes for that language and the marking its position takes, obligatory, optional or none; monotonicity is read over pairs of acceptable rows of one language and role in the product order of the parameters §9.4 separates, person, animacy proper and the degree of definiteness, a parameter the row does not cite constraining nothing. A row whose marking the book attributes to a parameter beside the hierarchies, Polish's male-human cut, carries a parameter feature and has no marking.
The departures from the paradigm case are per-language parameters: which positions a language doubles, whether it demotes further than required, and whether it has a passive, the last consulted only by the rival analysis and stated by the book for Turkish, French and Hungarian; Sanskrit's and Kannada's passives are textbook facts.
The prototype's predictions are stated through the bias of a construction type, agent-linked or patient-linked, which the book gives as the explanation; a control complement, the Chukchi infinitive construction, favours S with A on the book's account, and coordination has no bias.

TODO #

The book's four-grade definiteness continuum, complete identifiability > definite superset > discourse relevance > neither, is coarsened onto the substrate's scale, which merges the middle two grades; Persian (23) and Turkish (25) are therefore indefinite specific.
The book says Sanskrit has no dative causee at all, a position the hierarchy skips; the parameters record only doubling and further demotion, and no row tests the gap.
The doubling universal of §8.2, that the possibilities for doubling increase down the hierarchy, needs more languages than the book's rows give.
Chapter 9's animacy hierarchy is prose here: the Navajo yi- ~ bi- alternation, Chukchi number and the Tangut and Southern Tiwa agreement data would need agreement substrate.
The Russian numeral continuum of §5.2 and Greenberg's Universal 38 on the unmarkedness of the S case (§6.1) are not stated.

Comrie (1989): Language Universals and Linguistic Typology #

Implementation notes #

TODO #

References #

The discriminatory function of case (§6.1) #

The rows #

Differential marking and the natural information flow (§6.2) #

The causee hierarchy (§8.2) #

Compactness and directness (§8.1) #

Subject as a prototype (chapter 5) #