Haspelmath (2021): Role-reference associations and the explanation of argument coding splits #

Haspelmath's role-reference association universal (Universal 1, (5)) says that deviations from the usual associations of role rank and referential prominence are coded by longer grammatical forms. RoleReferenceUniversal states it for any usualness relation on situations. Its single-argument instance is SingleArgumentUniversal (Universal 3, whose role instances are Universals 4, 6, 7 and 8), and its scenario instance over Clause.Scenario is ScenarioUniversal (Universal 5, which yields Universals 9b, 10 and 11). Universal 12 reads Universal 1 with the usage rate of the longer alternant of an alternation as the coding, and Universals 13 and 14 are its scenario instances. The explanation of §11.2 is roleReferenceUniversal_of_formFrequency, by which Universal 2 and the form-frequency correspondence universal (68) of Haspelmath's companion paper entail Universal 1.

The coding splits the paper cites are coding-length functions on the prominence scales of (8), checked against the universals and against the paper's own examples (Reproduces), together with the anti-efficient language of §11.3 that the universals exclude.

Implementation notes #

The scales are the binary versions of (8), which are the two-element bounded orders (IsSimpleOrder). On them Universal 5 says exactly that coding is antitone in the order of Clause.Scenario, never shorter as the higher-ranked argument's prominence falls or the lower-ranked one's rises (scenarioUniversal_iff_antitone), which is what the paper uses tacitly for the monadic splits of fn. 18. The binary scales are load-bearing: on a scale with three levels a monadic scenario split obeys Universal 5 only if it is constant (ScenarioUniversal.comp_high_le), since a balanced scenario of two low values is coded no longer than an upstream one and no shorter than a downstream one. Kolyma Yukaghir's and Greek's rules on the ternary person scale are antitone and still violate Universal 5 (yukaghirP_greekT_ternary_antitone, yukaghirP_ternary_violates, greekT_ternary_violates). The ternary scales appear only where the paper's rule refers to them: PersonRank for Kashmiri, Animacy for Sardinian, Spanish, Fore and Awtuw, Nominality for Baule.
Universal 1's qualification to asymmetric coding is dropped: with coding lengths as the coding, a symmetric split has equal lengths and satisfies the universal vacuously, as Samoan's equal-length ergative alternants do (fn. 14). Its "tends to" is read as exceptionless, an idealization the paper itself qualifies. Universal 3's restriction to flagging (fn. 6) is not encoded, and the paper applies "longer coding" to Georgian (34) and Makassarese (56) itself; the index-based person-role constraint 9a (41) is omitted.
BinaryGivenness is read as the discourse-given ~ discourse-new scale of (8b), which the paper also calls topicality; Persian's topical indefinite P (19c) counts as given.
Coding length is an ℕ count of special coding elements in the cited systems; the universals are stated over any preorder, and Universal 12 is Universal 5 read with the usage rate of the longer alternant as the coding.
Universal 2 is a claim about discourse frequencies and enters as the hypothesis UsualIsFrequent of the reduction rather than as data. It is read within a role, across prominence values (a definite P is rarer than an indefinite P), which is what (68) opposes and what the reduction needs; (6) and the Appendix compare across roles at a fixed value (a definite A is commoner than a definite P), a weaker claim.
The example rows carry the coding an example shows (codingShown), so a strict split reproduces a row when the coding shown is acceptable exactly when it is the coding the split requires, and a splitting alternation when the construction shown is available in the row's scenario.

TODO #

Makassarese (56c), the applicative on a person-form T, is the nominality projection of a rule over definiteness and nominality together; the two projections are checked separately.
lummiPassive's availability whenever A is a full nominal is inferred from (64c) and (65b); the paper states only that the passive is obligatory in upstream scenarios and optional in N > N. UNVERIFIED against the source grammar.

References #

Referential prominence scales (8) #

Each scale is a linear order whose greater element is the more prominent. The person scale of (8a), locuphoric above aliophoric, is Person.Class; the definiteness scale of (8b) without its optional specific-indefinite level is Reference.Definiteness, with Eastern Khanty's specific P (37) read at the definite end; the givenness scale is BinaryGivenness; and the focus scale of (8b), background above focus, is the dual of focus marking, Focus.Markᵒᵈ, its more prominent end the unfocused one.

Haspelmath (2021): Role-reference associations and the explanation of argument coding splits #

Implementation notes #

TODO #

References #

Referential prominence scales (8) #

Cutoff splits #

Scenarios (10), (11) #

Universal 1 and its instances #

Cutoff splits #

Single-argument splits (§4, §5) #

Scenario splits (§6, §7) #

Relative scenario splits (§8) #

Verbal voice coding (§9) #

Splitting alternations (§10.2) #

The anti-efficient language (§11.3) #

The paper's examples #