Adamson 2024: Gender Assignment Is Local @cite{adamson-2024} #
@cite{adamson-2024} "Gender Assignment Is Local: On the Relation between Grammatical Gender and Inalienable Possession." Language 100(2): 218–264.
Core claim #
The Gender Locality Hypothesis (GLH): gender features on n must be valued only within nP. This restricts the conditioning factors for gender assignment to elements extremely local to the noun.
Key consequence for possession #
Inalienable possessors are introduced nP-internally (specifier of n with selectional feature {D}, following Myler 2016), while alienable possessors are introduced outside nP (specifier of PossP). The GLH therefore predicts:
- Inalienable possession CAN affect gender (nP-internal)
- Alienable possession CANNOT affect gender (outside nP)
This asymmetry is confirmed in four unrelated languages:
- Teop (Austronesian, Oceanic; §3.1): body-part nouns combine with two different n heads — n_{body-part{D}} bearing u[+ANIM] yields gender I when iPossessed; n_{alienator} (plain) yields gender II
- Jarawara (Arawan; §3.2): iPossessable roots are licensed only by plain n (feminine = unmarked in the [±MASC] system)
- Yanyuwa (Western Pama-Nyungan; §4.1): unvalued gender on n is valued by Probe-Goal agreement with the iPossessor DP
- Coastal Marind (Anim; §4.2): igih 'name' and nanVh 'face' inherit possessor's gender via agreement
Predictions beyond possession (§5) #
The GLH predicts that number features on Num (high number) cannot interact with gender, while number on n (low/derivational number) can. Features introduced on D (definiteness), T (tense), etc. are all outside nP and cannot affect gender assignment.
Connection to Linglib #
This module uses types from Theories/Morphology/DM/NominalStructure.lean
(the GLH, NominalPosition, PossessionType), CatHead and PhiBundle
from Theories/Morphology/DM/Categorizer.lean (@cite{kramer-2015}),
VocabItem from Theories/Morphology/DM/VocabularyInsertion.lean,
and Fragment data from Fragments/Teop/Nouns.lean and
Fragments/Jarawara/PossessedNouns.lean.
An n head with selectsD licenses an iPossessor in Spec,nP.
This connects CatHead.selectsD to the GLH: the {D} feature
places the possessor nP-internally, making it gender-relevant.
Gender features live on the nominal categorizer (n), as established by @cite{kramer-2015} and confirmed by @cite{adamson-2024}.
The ANIM-dimension types also carry gender features on n.
The GLH targets the nominal categorizer specifically. Verbal and adjectival categorizers do not host gender features.
Teop body-part n: bears u[+ANIM] and selectional feature {D}. When a body-part root combines with this n, the {D} feature creates a specifier position for an iPossessor DP. The u[+ANIM] feature results in gender I (animate article a).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Teop alienator n: plain n with no gender feature and no iPossessor.
Instances For
Determine gender from the n head's feature content. If n has any [ANIM]-dimension gender feature → gender I; otherwise → gender II.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Body-part nouns switch gender because they combine with two different n heads.
The body-part n licenses an iPossessor (has {D}); the alienator n does not.
The Teop gender alternation is consistent with the GLH.
The Teop body-part n uses the ANIM dimension, not the FEM dimension.
Teop Vocabulary Insertion #
Teop article VI rules, ordered by specificity.
Equations
- One or more equations did not get rendered due to their size.
Instances For
End-to-end: body-part root + n_{body-part{D}} → gender I → article a.
End-to-end: body-part root + n_{alienator} → gender II → article o.
Bridge to Fragment Data #
The study-level `teopGenderFromN` agrees with the Fragment-level
`iPossessedGender` for body-part nouns: both predict gender I
when iPossessed, gender II when free.
Five Teop Predictions (@cite{adamson-2024} §3.1) #
The two-n analysis generates five testable predictions (p.234–235):
Prediction 1: aPossessed body parts → gender II. When a body-part root combines with the alienator n (aPossession), the result is gender II, not gender I.
Prediction 2: the iPossessor's own gender is immaterial. The body-part n bears u[+ANIM] regardless of the possessor's features. This is possessee gender (determined by WHETHER iPossessed), not inherited gender (determined by possessor's gender value).
Prediction 3: the alienator n has no gender feature and no {D}. aPossession is mediated by PossP, not Spec,nP.
Prediction 4: any noun combining with n_{body-part{D}} gets gender I, not just canonical body parts. Relational nouns with the same structural profile (orientation terms, place nouns) also show the alternation.
Prediction 5: kinship nouns use the alienator n (aPossession), so they are always gender II regardless of possession.
Jarawara gender from the n head's feature content.
Equations
Instances For
The n head for Jarawara iPossessable nouns: has {D} (licenses iPossessor
in Spec,nP) but no gender feature (feminine = unmarked in the [±MASC]
system). This is distinct from CatHead.n_plain (which has selectsD = false).
Instances For
iPossessable roots in Jarawara are feminine: their n has no marked gender.
iPossessable n in Jarawara has {D} — by construction via CatHead.iPoss.
Masculine nouns in Jarawara bear [+MASC] on n.
Jarawara impoverishment (@cite{adamson-2024} ex. 63) #
Two separate impoverishment rules delete [MASC] in different contexts:
- [MASC] → ∅ / [PL]
- [MASC] → ∅ / [PARTICIPANT]
Impoverishment rule 1: [MASC] → ∅ / [PL].
Equations
- One or more equations did not get rendered due to their size.
Instances For
Impoverishment rule 2: [MASC] → ∅ / [PARTICIPANT].
Equations
- One or more equations did not get rendered due to their size.
Instances For
Both rules from ex. 63.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Impoverishment deletes [MASC] when [PL] is active.
Impoverishment deletes [MASC] when [PARTICIPANT] is active.
Impoverishment does not apply when context is inactive.
Bridge to Fragment Data #
The 175 iPossessable nouns in 12 semantic classes from `Fragments.Jarawara`
are drawn from the upper tiers of the inalienability hierarchy,
confirming the cross-linguistic prediction.
Inherited gender via Probe-Goal agreement #
@cite{adamson-2024} §4: in Yanyuwa and Coastal Marind, a small class of iPossessed nouns (igih 'name', nanVh 'face' in Coastal Marind; body parts and 'name' in Yanyuwa) "inherit" the gender of their iPossessor.
In Minimalist terms: the nominalizing head n has an unvalued gender feature. Because the iPossessor DP is in Spec,nP (nP-internal), Probe-Goal Agree can copy the possessor's valued gender onto n. The GLH permits this because the goal (iPossessor) is within nP.
A gender-inheriting noun: the n head bears an unvalued gender probe that is valued by Agree with the iPossessor DP's gender.
- rootGloss : String
The root (e.g., √IGIH 'name', √NANVH 'face').
- selectsD : Bool
The n head has {D} (selects an iPossessor).
- hasUnvaluedGender : Bool
The n head has an unvalued gender feature (probe).
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
- Phenomena.Morphology.Studies.Adamson2024.instBEqInheritedGenderNoun.beq x✝¹ x✝ = false
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
The n head for an inherited-gender noun: has {D} and an unvalued
gender probe. The probe is dimension-agnostic — it has no pre-specified
dimension or polarity. Its value (including dimension) comes entirely
from the iPossessor DP via Probe-Goal Agree (@cite{adamson-2024} (90)).
We represent this as phi := {} (no valued gender on n itself).
Instances For
Yanyuwa: seven gender classes (Kirton 1971a,b).
FEMALE, MALE, FEMININE (nonhuman female), MASCULINE (nonhuman male), FOOD, ARBOREAL, ABSTRACT. Body parts and 'name' take possessor prefixes expressing the possessor's φ-features.
- female : YanyuwaGender
- male : YanyuwaGender
- feminine : YanyuwaGender
- masculine : YanyuwaGender
- food : YanyuwaGender
- arboreal : YanyuwaGender
- abstract : YanyuwaGender
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
All seven Yanyuwa gender values.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Coastal Marind: four genders (Olsson 2017).
I (human men, e.g., yasti 'old man'), II (human women + animals, e.g., gomna 'male pig'), III (some inanimates, e.g., aliki 'river'), IV (other inanimates, e.g., himbu 'feathered headdress').
- gI : CoastalMarindGender
- gII : CoastalMarindGender
- gIII : CoastalMarindGender
- gIV : CoastalMarindGender
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
All four Coastal Marind gender values.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Coastal Marind inherited-gender nouns (Olsson 2017:187).
Equations
Instances For
Both inheriting nouns have {D} and unvalued gender — prerequisites for Probe-Goal agreement with the iPossessor.
Inherited gender is consistent with the GLH: the possessor whose gender is inherited occupies Spec,nP (nP-internal).
Bridge: Kramer's n-types and WALS gender counts #
@cite{kramer-2015} Ch 3: for a single gender dimension [±VAL], there are four types of n: i[+VAL], i[−VAL], plain, u[+VAL]. The fourth combination u[−VAL] is the unmarked default (plain n).
The maximum number of surface genders from one dimension is 3: the positive value, the negative value, and the default (plain). Two-gender systems arise when only one marked value + plain are attested.
Count distinct surface genders from a set of n heads. Two n heads produce the same surface gender iff they have the same gender feature content (ignoring interpretability, which is only visible at LF vs PF).
Equations
- Phenomena.Morphology.Studies.Adamson2024.surfaceGenderClass nh = Option.map (fun (x : Morphology.DM.GenderFeature) => x.val) nh.phi.gender
Instances For
The four Amharic n-types yield exactly 3 surface genders: [+FEM], [−FEM], and ∅ (plain). Both i[+FEM] and u[+FEM] map to the same surface class [+FEM].
A two-gender system (e.g., Jarawara [±MASC]) uses only two n types: marked (u[+MASC]) and plain.
The Teop two-gender system uses the ANIM dimension.
Every n-head used for inalienable possession must have selectsD = true.
Without this, the n-head cannot license an iPossessor in Spec,nP, and the
semantic pipeline (catHeadSemanticType) will compute sortal instead of
relational. This invariant was violated before 0.229.208 (Jarawara used
CatHead.n_plain which has selectsD = false).
All iPossessable n-heads across all four languages have selectsD = true. Regression test: adding a new iPossessable n-head? Add it to the disjunction here. If it fails, the n-head is missing {D}.
Two derivation pipelines from a single n-head #
A CatHead determines two things in parallel:
┌──→ gender ──→ article form (PF pipeline)
CatHead ──┤
└──→ NSemanticType ──→ can take possessor? (semantic pipeline)
The PF pipeline genuinely threads: teopGenderFromN computes the gender,
which feeds into vocabularyInsert as the article context. The semantic
pipeline threads similarly: catHeadSemanticType computes the semantic
type, which feeds into .toBarker.canTakePossessor.
The non-trivial claim is that these two pipelines produce correlated
outputs: gender I co-occurs with relational semantics (possessor slot),
and gender II co-occurs with non-relational (no possessor slot). The
correlation is structural — both paths read selectsD from the same
n-head.
The PF derivation pipeline: n-head → gender → article. Gender is an intermediate value computed from φ-features, then fed into VI as part of the article context.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The semantic derivation pipeline: n-head → semantic type → possessor capability. The semantic type is an intermediate value, fed into Barker's type classification to determine whether the noun can directly take a possessor. Generic over any CatHead — not Teop-specific despite the examples below.
Equations
- Phenomena.Morphology.Studies.Adamson2024.canTakePossessorSem nh mediatesAPoss = (Morphology.DM.CategorizerSemantics.catHeadSemanticType nh mediatesAPoss).toBarker.canTakePossessor
Instances For
The two pipelines produce correlated results: the PF pipeline yields gender I exactly when the semantic pipeline yields possessor capability.
This is the Adamson–Barker correspondence. The gender alternation (gender I vs II) and the semantic type alternation (relational vs non-relational) are not independent — they are both downstream consequences of the same structural feature (selectsD on n).
Stated as a Bool identity: "is gender I" = "can take possessor".
Jarawara: PF pipeline via manoForm #
Jarawara doesn't have articles, but it DOES have a PF pipeline: possessor
features → impoverishment → MARKED feature check → possessed noun form
(mano vs mani). This is the manoForm function from Fragments.Jarawara.
The semantic pipeline parallels Teop: the iPossessable n has {D} → relational semantic type → can take possessor. The gender is feminine because n has no marked gender feature, not because it lacks {D}.
Jarawara PF pipeline: possessor features → impoverishment → MARKED feature check → possessed form (mano/mani). The possessor's features thread through each stage.
Equations
Instances For
Jarawara iPossessable n has {D} → relational semantic type. The n head licenses an iPossessor AND yields relational (π) semantics.
Jarawara PF–semantic correlation: iPossessable n yields feminine gender (no marked feature) AND relational semantics (has {D}). The correlation shows that Jarawara iPossessable nouns CAN take possessors despite being feminine — femininity reflects the absence of a gender feature, not the absence of a possessor slot.
Inherited gender: the GLH contrast #
For Yanyuwa and Coastal Marind, the n-head has {D} → relational semantics. The key GLH prediction is a contrast: iPossessors (specN) can affect gender, aPossessors (specPoss) cannot. The contrast is captured by the GLH function applied to two different positions — not a chain, just two evaluations of the same predicate on different inputs.
The inherited-gender n head has {D} → relational semantic type. Gender is unvalued on n and comes from the iPossessor via Agree.
The GLH contrast: iPossessors can affect gender, aPossessors cannot.
This is not a derivation — it's two evaluations of genderLocalityHypothesis
on the two possessor positions. The function does the work.
Beyond possession: number position and gender #
@cite{adamson-2024} §5.1 extends the GLH beyond possession: if gender features sit on n, then OTHER features on n should also interact with gender. Number features on n (low/derivational number) are within nP and can interact with gender; number features on Num (high/inflectional) are outside nP and cannot.
Standard Italian confirms this: the -a plural class (braccio → braccia 'arm → arms') changes gender from masculine to feminine. These are low-number plurals (number on n). Regular -i plurals (libro → libri) preserve gender — they involve high number (on Num).
The same function (genderLocalityHypothesis) that predicts possession–gender
interaction in Teop and Jarawara also predicts number–gender interaction in
Italian. The GLH is a single principle applied to different feature types.
The Italian data confirms the GLH prediction: gender changes in the
plural track the number position. Verified over the full noun inventory.
The pipeline (numberGenderPipeline) composes
PluralClass → NumberPosition → NominalPosition → GLH.
Cross-linguistic convergence: the -a plural class is dominated by body parts (6 of 9). Body parts drive gender interaction in ALL four languages examined:
- Teop: body parts switch gender I/II with iPossession
- Jarawara: body parts are always iPossessable (feminine)
- Yanyuwa/Coastal Marind: body parts inherit possessor's gender
- Italian: body parts show the -a plural gender alternation