Differential Indexing #
[Ais03] [Jus24] [Has19] [Sie04] [Pre14] [Has21]
Formalizes the typological survey of Just, E. (2024), A structural and
functional comparison of differential A and P indexing (Linguistics 62(2):
295–321), and connects it to [Ais03]'s DOM profiles
(Features/Prominence), [Pre14]'s φ-geometry
(Syntax/Minimalist/Phi/), and the Kaqchikel, Basque, Georgian, and
Hungarian fragments.
Flagging = case morphology on the NP; indexing = verbal agreement/cross-referencing ([Has19], Just §2). The same prominence scales (person, animacy, definiteness) govern both, with opposite polarity by role (Just §4.2, p. 311): P indexing targets prominent Ps, A indexing targets non-prominent As. Both follow from a single principle (Just §6, p. 315): prominent arguments are indexed more readily regardless of role; A defaults prominent, P does not.
Main declarations #
IndexingFragment: per-language differential-indexing profile, extendingDifferentialMarkingProfile(channel.indexing)pIndexingLanguages,aIndexingLanguages: Just (2024) Tables 1–2mirror_image_universal: every language targets the role-appropriate end of each conditioning scaleperson_dominates_P/person_dominates_A: person is the most frequent conditioning factor for both rolespersonLevel_matches_participant: Just's SAP/3rd split is [Pre14]'s [±participant]basque_fragment_matches_survey,georgian_fragment_matches_survey,hungarian_conjugation_split: fragment paradigms ground the survey rows
Person prominence for differential indexing.
The person scale for indexing is a binary split between speech act participants (SAP: 1st/2nd person) and non-participants (3rd person). This mirrors [Pre14]'s [±participant] feature decomposition.
This is coarser than Person (1st > 2nd > 3rd),
which is needed for scenario splits. The binary split suffices for
indexing because indexing does not distinguish 1st from 2nd.
- sap : IndexingPersonLevel
Speech act participants: 1st and 2nd person
- third : IndexingPersonLevel
Non-participants: 3rd person
Instances For
Equations
- Aissen2003.instDecidableEqIndexingPersonLevel x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Rank on the indexing person scale: SAP (1) > 3rd (0).
Instances For
All indexing person levels.
Equations
Instances For
A differential indexing fragment for a single language.
Extends DifferentialMarkingProfile with language metadata and
per-dimension marking predicates. The 2D marks predicate is computed
from animacyIndexed and definitenessIndexed by the mk' constructor,
and the channel is always .indexing.
For P indexing: true at a level means "P arguments at this level
ARE indexed." The expectation (Just §4.2) is that P indexing targets
the prominent end of each scale.
For A indexing: true at a level means "A arguments at this level
ARE indexed." The expectation is that A indexing targets the
non-prominent end.
- name : String
- marks : AnimacyLevel → DefinitenessLevel → Bool
- iso639 : String
ISO 639-3 code
- family : String
Language family
- personIndexed : IndexingPersonLevel → Bool
Which person levels trigger indexing
- animacyIndexed : Features.Prominence.AnimacyLevel → Bool
Which animacy levels trigger indexing
- definitenessIndexed : Features.Prominence.DefinitenessLevel → Bool
Which definiteness levels trigger indexing
Instances For
Smart constructor for IndexingFragment.
Computes the 2D marks predicate as the product of
animacyIndexed and definitenessIndexed, and sets channel :=.indexing.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A dimension is conditioning if the marking predicate is non-uniform: some levels are indexed and some are not.
Equations
Instances For
Equations
Instances For
Equations
Instances For
At least one dimension conditions the indexing (i.e., is non-uniform).
Equations
- f.isDifferential = (f.personConditioned || f.animacyConditioned || f.definitenessConditioned)
Instances For
P indexing has correct polarity if it targets the PROMINENT end: SAP over 3rd, human over inanimate, definite over nonspecific.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A indexing has correct polarity if it targets the NON-PROMINENT end: 3rd over SAP, inanimate over human, nonspecific over definite.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Role-appropriate polarity check. T behaves like P (targets prominent), R behaves like A (targets non-prominent). S is vacuously correct.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Languages where the P argument is differentially indexed. Prominent Ps (SAP, human, definite) are MORE likely to be indexed.
Marking predicates encode the actual pattern per scale. For scales that
are not conditioning, the predicate is uniformly `true` (all levels
indexed). For conditioning scales, the predicate marks the prominent
end as `true` and the non-prominent end as `false`.
Source references are from [just-2024].
Abkhaz (NW Caucasian): P indexed only for SAP.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Amharic (Semitic): P indexed for SAP and definite objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Basque (Isolate): object agreement only for SAP objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Georgian (Kartvelian): P agreement conditioned by person — indirect objects (dative) are indexed for SAP only.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Hungarian (Uralic): definite conjugation triggered by definite objects;
indefinite conjugation for indefinite objects.
See also Hungarian.Predicates for the conjugation split.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Kagulu (Bantu): object marker for animate+ objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
KiNzadi (Bantu): P indexed for SAP only.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Koorete (Omotic): P indexed for SAP only.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Maltese (Semitic): definite object agreement — verb agrees with definite objects via suffixed object markers (Just & Čéplö 2022).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Nkore-Kiga (Bantu): object marker for SAP objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Romanian (Romance): clitic doubling conditioned by person, animacy, and definiteness. SAP + human + definite = doubled.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Somali (Cushitic): P indexed for SAP — full paradigm for SAP objects, reduced for 3rd person. Focus/topicality also plays a role but is not captured in the prominence grid.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Swahili (Bantu): object marker obligatory for human objects, optional/ absent for non-human.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Teiwa (Trans-New Guinea): P indexed for animate objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Welsh (Celtic): synthetic agreement only with pronominal (SAP) objects; analytic (no agreement) with 3rd person full NPs.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Zulu (Bantu): object marker for animate objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Spoken Arabic varieties (Semitic): P indexed for SAP only — object suffixes restricted to pronominal objects.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Languages where the A argument is differentially indexed. Non-prominent As (3rd person, inanimate, indefinite) are MORE likely to be indexed. The polarity is REVERSED relative to P indexing.
Source references from [just-2024].
Jamsay (Dogon): A indexed only for 3rd person agents — SAP agents are not cross-referenced on the verb.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Kharia (Munda): A indexed for 3rd person agents.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Mundari (Munda): A indexed for 3rd person agents.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Juang (Munda): A indexed for 3rd person agents.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Anywa (Nilotic): A indexed for 3rd person agents only — SAP agents are not cross-referenced.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Reyesano (Tacanan): A indexed for 3rd person and non-human agents.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Eastern Mansi (Uralic): A indexed for indefinite/non-topical agents.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All differential P indexing languages in the sample.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All differential A indexing languages in the sample.
Equations
Instances For
All differential indexing languages.
Equations
Instances For
All P indexing languages have role P.
All A indexing languages have role A.
All profiles are genuinely differential (at least one conditioning factor is non-uniform). Derived from the marking predicates.
The collections below are COMPUTED from the marking predicates via
personConditioned, animacyConditioned, definitenessConditioned —
if a language's marking predicate changes, they update automatically.
P indexing languages conditioned by person (derived).
Equations
- Aissen2003.pPersonConditioned = List.filter (fun (x : Aissen2003.IndexingFragment) => x.personConditioned) Aissen2003.pIndexingLanguages
Instances For
P indexing languages conditioned by animacy (derived).
Equations
- Aissen2003.pAnimacyConditioned = List.filter (fun (x : Aissen2003.IndexingFragment) => x.animacyConditioned) Aissen2003.pIndexingLanguages
Instances For
P indexing languages conditioned by definiteness (derived).
Equations
- Aissen2003.pDefinitenessConditioned = List.filter (fun (x : Aissen2003.IndexingFragment) => x.definitenessConditioned) Aissen2003.pIndexingLanguages
Instances For
A indexing languages conditioned by person (derived).
Equations
- Aissen2003.aPersonConditioned = List.filter (fun (x : Aissen2003.IndexingFragment) => x.personConditioned) Aissen2003.aIndexingLanguages
Instances For
Person is the most common conditioning factor for P indexing: more P-indexing languages are person-conditioned than animacy- or definiteness-conditioned.
Person is the most common conditioning factor for A indexing: more A-indexing languages are person-conditioned than animacy- or definiteness-conditioned.
"The very same referential properties condition both differential P and differential A indexing."
Person conditions both P and A indexing.
Animacy conditions both P and A indexing.
Definiteness conditions both P and A indexing.
"The directions in which these scales operate form a mirror image: indexing targets prominent P arguments on the one hand and non-prominent A arguments on the other."
All P indexing languages have correct polarity: they target the prominent end of each conditioning scale.
All A indexing languages have correct polarity: they target the non-prominent end of each conditioning scale.
The mirror image holds universally across the sample: every language has role-appropriate polarity.
For languages where animacy and/or definiteness condition indexing,
we verify monotonicity of the inherited DifferentialMarkingProfile
on the 2D grid. The marks predicate is derived from the fragment's
animacyIndexed and definitenessIndexed — no separate stipulation.
All animacy/definiteness-conditioned profiles are monotone.
Swahili P indexing depends only on animacy (definiteness is irrelevant).
Kagulu P indexing depends only on animacy.
Hungarian P indexing depends only on definiteness (animacy is irrelevant).
Eastern Mansi A indexing depends only on definiteness.
Bantu languages in the sample all show P indexing.
Munda languages in the sample all show A indexing.
"Prominent arguments, be it A or P (or probably any other role), tend to be indexed more readily than arguments which are low in identifiability, animacy or topicality."
The mirror image is NOT a coincidence — it follows from a single
principle (indexing tracks prominent referents) combined with different
default prominence per role (A defaults high, P defaults low).
For every person-conditioned P-indexing language, SAP is indexed (= prominent P gets indexed).
For every person-conditioned A-indexing language, 3rd person is indexed (= non-prominent A gets indexed, because A's default is prominent).
The unified principle predicts that the COMPLEMENT of "indexed" differs by role: P-indexing excludes 3rd person, A-indexing excludes SAP. Both follow from "index the prominent referent" + role defaults.
DOMProfile is an abbreviation for DifferentialMarkingProfile
(specialized to role P + channel flagging), and IndexingFragment
extends DifferentialMarkingProfile (with channel = .indexing).
Both DOM profiles and indexing fragments inherit all DMP infrastructure
(monotonicity, dimensionality, cutoff constructors, mirror image)
directly — no conversion or bridge theorems needed.
Just's binary person split (SAP vs 3rd) is exactly Preminger's [±participant] feature decomposition.
Map a Person to Just's IndexingPersonLevel. 1st/2nd → SAP, 3rd → third.
Equations
- Aissen2003.personToLevel Person.first = Aissen2003.IndexingPersonLevel.sap
- Aissen2003.personToLevel Person.firstInclusive = Aissen2003.IndexingPersonLevel.sap
- Aissen2003.personToLevel Person.firstExclusive = Aissen2003.IndexingPersonLevel.sap
- Aissen2003.personToLevel Person.second = Aissen2003.IndexingPersonLevel.sap
- Aissen2003.personToLevel Person.third = Aissen2003.IndexingPersonLevel.third
- Aissen2003.personToLevel Person.zero = Aissen2003.IndexingPersonLevel.third
Instances For
personToLevel agrees with decomposePerson on the participant split: SAP ↔ [+participant], third ↔ [−participant].
SAP has higher prominence rank than 3rd, just as [+participant] gives higher probe resolution rank.
Kaqchikel indexes both A and P arguments uniformly across all person-number combinations. This is a NON-differential system: there is no prominence-based asymmetry in which arguments get indexed.
[just-2024] defines differential indexing against this kind of
baseline: a differential system is one where indexing depends on
prominence properties.
Kaqchikel indexes all three argument positions (agent, patient, intranS). This makes it non-differential: no prominence condition gates indexing.
Both A (agent) and P (patient) are indexed in Kaqchikel: agent via Set A on Voice/v, patient via Set B on Infl/T.
The A marker paradigm (Set A) and P marker paradigm (Set B) are distinct: every person-number combination gets a unique marker in each set (except 3SG which is ∅ in Set B).
Map Kaqchikel argument positions to Just's A/P roles. The absolutive collapse: S patterns with P, A stays distinct; ditransitive R/T default to P (consistent with absolutive grouping).
Equations
- Aissen2003.kaqArgToRole Features.Prominence.ArgumentRole.A = Features.Prominence.ArgumentRole.A
- Aissen2003.kaqArgToRole Features.Prominence.ArgumentRole.S = Features.Prominence.ArgumentRole.P
- Aissen2003.kaqArgToRole Features.Prominence.ArgumentRole.P = Features.Prominence.ArgumentRole.P
- Aissen2003.kaqArgToRole Features.Prominence.ArgumentRole.R = Features.Prominence.ArgumentRole.P
- Aissen2003.kaqArgToRole Features.Prominence.ArgumentRole.T = Features.Prominence.ArgumentRole.P
Instances For
Identity on A and P; the load-bearing structure is the S → P
collapse encoded in kaqArgToRole.
Ergative-absolutive alignment: A is distinguished (ERG) while P and S pattern together (ABS). This parallels Just's A/P split.
Person is the dominant conditioning factor for both P indexing and
A indexing (§10). The structural correlate is that the [participant]
probe (π⁰) takes priority over the [plural] probe (#⁰) under the
two-probe relativized-probing system [BR03] —
NOT a salience hierarchy. [Pre14] Ch. 7 explicitly
argues against direct salience-scale primitives; the rank
ordering below is a surface effect of probe priority, not a
hierarchy-as-grammatical-primitive. See
Studies/Preminger2014.lean for the
anti-hierarchy theorems.
Two-probe surface ranking [BR03]: [+participant] cells outrank [+plural,−participant] cells, which outrank 3SG. The typological frequency hierarchy (person > animacy > definiteness) parallels this — person features are both structurally privileged at the probe level and typologically dominant in indexing systems.
The Basque agreement fragment (Basque.Agreement) encodes
the same person-conditioned P indexing that [Jus24] reports.
We prove that the Fragment's pIsIndexed matches the survey data.
Basque Fragment's P indexing matches the Just survey: SAP → indexed,
3rd → not indexed, exactly as basque.personIndexed.
Basque Fragment confirms differential P indexing: some indexed, some not.
The Georgian agreement fragment (Georgian.Agreement) derives
P indexing from the presence of object agreement prefixes (m-, g-, gv-).
The indexed/not-indexed split aligns with SAP vs 3rd — same as the
Just survey data.
Georgian Fragment's P indexing matches the Just survey.
Georgian Fragment's P indexing is grounded in object prefix morphology: indexed iff the object paradigm realizes the cell. Not stipulated — follows from the data.
The Hungarian predicate fragment (Hungarian.Predicates)
models the definite/indefinite conjugation split. This IS Just's
differential P indexing by definiteness: the verb's agreement paradigm
changes depending on whether the object is definite.
The fragment's `formPastDef ≠ formPastIndef` encodes the same claim
as the Just survey entry `hungarian.definitenessConditioned`.
Hungarian verbs have distinct definite vs indefinite conjugation forms. This IS the morphological reflex of differential P indexing by definiteness.
Hungarian is definiteness-conditioned (derived from the marking predicate), confirming the Fragment's conjugation split.
Hungarian is NOT person-conditioned — all persons can trigger both conjugation types.