Adamson & Anagnostopoulou 2025 @cite{adamson-anagnostopoulou-2025} #
Gender Features and Coordination Resolution in Greek and Other Three-Gendered Languages: Implications for the Crosslinguistic Representation of Gender. Linguistic Inquiry (Early Access).
Core result #
Cross-linguistic variation in coordination resolution (Table 2) is derived from three interacting components, without stipulated defaults:
Feature geometry: language-specific hierarchies of privative gender features. In Greek, FEM is a dependent of MASC (= animate), so a feminine nominal also bears MASC. In Icelandic, MASC (= male) and FEM are independent under CLASS — neither entails the other.
Dual-feature system: interpretable features (iFs → LF) vs uninterpretable features (uFs → PF), linked by a redundancy rule that copies iFs to empty uF slots at Transfer.
Percolation + conversion: universal coordination resolution. iFs percolate from conjuncts to &P; shared iFs survive intersection (conversion); the Subset Principle selects the inflectional exponent.
Table 2 — resolution patterns derived from geometry:
| Language | Humans | Inanimates |
|---|---|---|
| Greek | MASC | NEUT |
| Icelandic | NEUT | NEUT |
| BCS | MASC | MASC |
Y-model architecture #
The derivation follows the Minimalist Y-model:
- Narrow syntax: nominals bear both iFs and uFs; percolation collects features at &P.
- Transfer: conversion intersects iFs; redundancy rule copies resolved iFs to empty uF slots.
- PF: Vocabulary Insertion (Subset Principle) maps uF sets to inflectional exponents.
- LF: Lexical Complementarity restricts feature interpretation (e.g., iMASC restricted to ⟦MASC⟧ − ⟦FEM⟧).
Formalization #
Resolution uses GenderResolution.resolve — the single compositional
endpoint — instantiated with GenderNode as the feature type. Each
prediction is a verified theorem.
Privative feature nodes in the gender geometry (§2.2).
@cite{adamson-anagnostopoulou-2025} modifies @cite{harley-ritter-2002}: features are organized in language-specific hierarchies where more specific features entail broader features. The same labels appear across languages, but their geometric arrangement — and hence their semantic content — varies.
Language-specific geometries:
- Greek (17): CLASS > MASC > FEM (linear chain)
- Icelandic (63): CLASS > {MASC, FEM} (independent siblings)
- BCS (74): CLASS > INDIV > {GRP, MASC > ANIM > FEM}
- cls : GenderNode
- masc : GenderNode
- fem : GenderNode
- indiv : GenderNode
- grp : GenderNode
- anim : GenderNode
Instances For
Equations
- AdamsonAnagnostopoulou2025.instDecidableEqGenderNode 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
Partial bridge from privative geometry nodes to DM gender dimensions.
.cls, .indiv, and .grp are structural organizing nodes with no
counterpart in the DM GenderDimension type (which tracks only
semantic gender dimensions).
Equations
- AdamsonAnagnostopoulou2025.GenderNode.masc.toGenderDimension = some Morphology.DM.GenderDimension.masc
- AdamsonAnagnostopoulou2025.GenderNode.fem.toGenderDimension = some Morphology.DM.GenderDimension.fem
- AdamsonAnagnostopoulou2025.GenderNode.anim.toGenderDimension = some Morphology.DM.GenderDimension.anim
- x✝.toGenderDimension = none
Instances For
Equations
- AdamsonAnagnostopoulou2025.instDecidableEqInfl 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
Bridge from VI inflection class to cross-linguistic surface gender.
Equations
Instances For
Greek vocabulary item schema (21).
| uF specification | Exponent |
|---|---|
| {FEM, MASC} | "feminine inflection" |
| {MASC} | "masculine inflection" |
| ∅ | "neuter inflection" |
The Subset Principle selects the most specific matching item.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Icelandic vocabulary schema — identical to Greek. The geometry difference, not the vocabulary, drives the divergent resolution patterns.
Instances For
BCS vocabulary item schema (75).
| uF specification | Exponent |
|---|---|
| {FEM, ANIM, MASC, INDIV} | "F inflection" |
| {ANIM, MASC, INDIV} | "M animate inflection" |
| {INDIV} | "M inanimate inflection" |
| ∅ | "N inflection" |
Simplified: FEM ∧ ANIM → F; INDIV present → M; else → N.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Greek feature geometry (17) #
CLASS > MASC > FEM (linear chain).
Feature interpretations (18):
- CLASS: λx. x is an entity
- MASC: λx. x is animate
- FEM: λx. x is a woman
Having FEM entails MASC; having MASC entails CLASS.
Lexical Complementarity (19) restricts: iMASC picks out ⟦MASC⟧ − ⟦FEM⟧
(animate non-women); iFEM picks out women; iCLASS picks out entities.
(25) Uniform humans resolve to their shared gender.
(22a) Mismatched humans → {CLASS, MASC} → masculine plural. Intersection: {CLASS,MASC,FEM} ∩ {CLASS,MASC} = {CLASS,MASC}. FEM is eliminated because only one conjunct bears it.
(22b, 40a-c) All inanimate mismatch combinations → {CLASS} → neuter. Percolation extracts only iCLASS from each conjunct (uFs excluded). This is NOT default insertion — it is the result of intersecting the iF sets, which contain only iCLASS for all inanimates.
Mismatch Resolution Hypothesis (24): no default feature insertion. All resolution outcomes have non-empty intersection (matching).
Fixed-gender human nominals (§3.2) #
*megalofiia* 'genius' is **grammatically** feminine (uFs) but its
**conceptual** gender (iFs) tracks the referent. Resolution operates
on iFs — crucial evidence for the Mismatch Resolution Hypothesis.
*thima* 'victim' is grammatically neuter (uF = {CLASS}) but
conceptually masculine/feminine depending on referent.
(36) megalofiia (male referent) + sister → masculine (M♂ + F♀ = M). Despite both being grammatically feminine, iF resolution yields {CLASS,MASC} ∩ {CLASS,MASC,FEM} = {CLASS,MASC} → MASC.
(37) thima (female referent) + her mother → feminine (N♀ + F♀ = F). Neuter noun's iFs are feminine (referent is female): {CLASS,MASC,FEM} ∩ {CLASS,MASC,FEM} = {CLASS,MASC,FEM} → FEM.
Humans + Inanimates (§3.5) #
When a human (percolating iFs) and an inanimate (percolating uFs)
are coordinated, PF realization succeeds only if the features map
to the same inflection class. Otherwise: PF crash → ineffability.
Y-model significance: the crash happens at **PF**, not in narrow
syntax. [H + I] coordination is syntactically well-formed; the
problem is exponence of the resolved features.
PF convergence test for [H + I] coordination. The human's iFs (→ uFs via redundancy rule at Transfer) must map to the same inflection as the inanimate's uFs.
Equations
- AdamsonAnagnostopoulou2025.gkHIConverges humanIFs inanimUFs = (AdamsonAnagnostopoulou2025.greekVI humanIFs == AdamsonAnagnostopoulou2025.greekVI inanimUFs)
Instances For
(54a) M♂ + M■ → grammatical: both map to MASC. kleftis 'thief' (M) + pinakas 'painting' (M).
(54b) F♀ + F■ → grammatical: both map to FEM. gineka 'woman' (F) + ombrela 'umbrella' (F).
(47) M♂ + N■ → ineffable: MASC ≠ NEUT → PF crash. kleftis 'thief' (M♂) + daxtilidi 'ring' (N■). Human iFs → MASC; inanimate uFs → NEUT. No single exponent.
(56) Fixed-gender F♂ + M■ → grammatical: iFs match. megalofiia 'genius' (iF = M♂) + pinakas 'painting' (uF = M■). Human's iFs = {CLASS,MASC} → MASC; inanimate's uFs = {CLASS,MASC} → MASC. PF converges despite different grammatical genders.
Fixed-gender humans + inanimates (§3.5) #
The paper's strongest evidence for the iF-based analysis. When a
**fixed-gender human** is coordinated with an inanimate, the PF
convergence depends on the human's **iFs** (conceptual gender),
not their **uFs** (grammatical gender).
(57a) thima (N♀ victim, male referent) + pinakas (M■) → MASC. iFs of victim (male) = {CLASS, MASC}; uFs of painting = {CLASS, MASC}. VI match: MASC = MASC → PF converges.
(57b) thima (N♀ victim, female referent) + fotografia (F■) → FEM. iFs of victim (female) = {CLASS, MASC, FEM}; uFs of picture = {CLASS, MASC, FEM}. VI match: FEM = FEM → PF converges.
(57a corollary) thima (N♀, male ref) + fotografia (F■) → PF crash. iFs of victim (male) = {CLASS, MASC} → VI → MASC. uFs of picture = {CLASS, MASC, FEM} → VI → FEM. MASC ≠ FEM → crash.
Uniform inanimate coordination (38a-c) #
When two inanimates share the same grammatical gender, resolved
agreement matches that gender. Under the paper's analysis, this
obtains when inanimates percolate **uFs** (the alternative
derivation in (39)): the &P has a singleton uF set, and PF
realization succeeds with the shared exponent.
(38a) F■ + F■ = F: fusta 'skirt' + bluza 'T-shirt'.
(38b) M■ + M■ = M: anaptiras 'lighter' + fakos 'torch'.
(38c) N■ + N■ = N: piruni 'fork' + kutali 'spoon'.
(2a, 58) Clausal subjects lack gender features entirely. No features to percolate → no vocabulary item matches → neuter (the elsewhere exponent, least specified).
Icelandic feature geometry (63) #
```
(i/u)CLASS
/ \
MASC FEM ← independent (no entailment)
```
Unlike Greek, FEM is NOT a dependent of MASC. The crucial difference:
MASC means 'male' (not 'animate'), so gender-mixed groups are excluded
from both MASC (not all male) and FEM (not all female).
Lexical Complementarity: no restriction between MASC and FEM
(neither is a subset of the other).
(60) Mismatched humans → {CLASS} → neuter. {CLASS,MASC} ∩ {CLASS,FEM} = {CLASS}. Because MASC and FEM are independent siblings, only CLASS survives intersection.
(59) Mismatched inanimates → {CLASS} → neuter. frægð 'fame' (F) + frami 'success' (M) → neuter plural. All Icelandic inanimates share iFs = {iCLASS} regardless of grammatical gender — uFs (uFEM, uMASC) are excluded from resolution.
The geometry contrast: same labels, different geometry, different outcome. Greek {CLASS,MASC,FEM} ∩ {CLASS,MASC} = some {CLASS,MASC}. Icelandic {CLASS,FEM} ∩ {CLASS,MASC} = some {CLASS}. Same mechanism, different input → different result.
BCS feature geometry (74) #
```
CLASS
|
INDIV
/ \
GRP MASC
|
ANIM
|
FEM
```
Key differences from Greek and Icelandic:
1. MASC is under INDIV (not directly under CLASS)
2. Neuter ≈ mass (no INDIV) → can't be counted or coordinated as count
3. All coordinatable nominals have INDIV → resolved features include
at least {INDIV} → vocabulary maps to masculine
Vocabulary consequence: masculine is the "default" for coordination
not by stipulation but because INDIV (required for plural) maps to
masculine via the Subset Principle.
(68) Mismatched humans → {CLASS, INDIV, MASC, ANIM} → masculine. {CLASS,INDIV,MASC,ANIM,FEM} ∩ {CLASS,INDIV,MASC,ANIM} = {CLASS,INDIV,MASC,ANIM}.
Mismatched M + F inanimates → {CLASS, INDIV, MASC} → masculine. Both conjuncts share {CLASS, INDIV, MASC} as iFs; ANIM + FEM on the feminine noun are uFs, excluded from resolution.
(69) Mismatched N + F inanimates → masculine. znanje 'knowledge' (N) + intuicija 'intuition' (F). Neuter is mass (iFs = {CLASS}); feminine inanimate has iFs = {CLASS, INDIV, MASC}. Intersection = {CLASS}. But coordination introduces GRP (entailing INDIV), so the &P bears {CLASS, INDIV} at minimum → Subset Principle yields masculine.
NB: The formalization models this by checking that the gender resolution itself yields {CLASS}, and the coordinate structure's INDIV (from GRP/plural) independently ensures masculine VI.
After coordination introduces INDIV, the combined features ({CLASS} from resolution + {INDIV} from GRP) map to masculine.
(70) Even matched neuters → masculine when coordinated. selo 'village' (N) + brdo 'hill' (N). Individual neuter DPs are mass (iFs = {CLASS}). Resolution: {CLASS} ∩ {CLASS} = {CLASS}. But coordination introduces GRP (entailing INDIV). The combined {CLASS, INDIV} at &P maps to masculine via VI — INDIV is present → masculine.
Table 2 verified: all six cells derived from geometry + intersection.
Redundancy rule (13): copy iF values to empty uF slots at Transfer. If uFs are already filled (arbitrary gender), the rule does not apply.
Y-model: this operation occurs at Transfer, the boundary between narrow syntax and the PF/LF interfaces.
Equations
- AdamsonAnagnostopoulou2025.redundancyRule iFs uFs = if uFs.isEmpty = true then iFs else uFs
Instances For
Human feminine: uFs empty → redundancy fills from iFs.
Arbitrary feminine: uFs already filled → preserved.
The implicational feature hierarchy (CLASS > MASC > FEM) predicts the absence of ABA syncretism patterns (fn. 19). Neuter and feminine can never be syncretic to the exclusion of masculine — since the masculine feature set {MASC} is a proper subset of feminine {FEM,MASC} and a proper superset of neuter ∅, any syncretism of N and F would also include M.
Greek inflection: M is always "between" N and F in feature specificity. This rules out N = F ≠ M syncretism patterns.
Geometry → outcome #
The paper's central thesis: cross-linguistic variation in resolution
follows from differences in feature geometry, not from different
resolution mechanisms or stipulated defaults. Same labels, same
mechanism, different geometry → different outcome.
We formalize each language's geometry as a function from a base
gender node to the full set of entailed iFs, then prove that
resolution outcomes follow from geometry alone.
Greek geometry (17): CLASS > MASC > FEM (linear chain). FEM entails MASC entails CLASS.
Equations
- One or more equations did not get rendered due to their size.
- AdamsonAnagnostopoulou2025.greekGeometry AdamsonAnagnostopoulou2025.GenderNode.cls = [{ value := AdamsonAnagnostopoulou2025.GenderNode.cls, interp := Minimalist.Interpretability.interpretable }]
- AdamsonAnagnostopoulou2025.greekGeometry x✝ = []
Instances For
Icelandic geometry (63): CLASS > {MASC, FEM} (independent siblings). Neither FEM nor MASC entails the other.
Equations
- One or more equations did not get rendered due to their size.
- AdamsonAnagnostopoulou2025.icelandicGeometry x✝ = []
Instances For
The linear chain geometry guarantees that mismatched human resolution retains MASC — because both conjuncts bear iMASC (FEM entails MASC).
The independent geometry means human mismatch loses both MASC and FEM — only CLASS survives intersection.
Geometry determines resolution outcome: same mechanism + same feature labels → different VI output, entirely from geometry.
The geometry functions reconstruct the noun data: Greek nouns are exactly the geometry applied to their most specific iF.
Entailment asymmetry: in the linear chain (Greek), FEM's iFs are a superset of MASC's iFs. In the independent geometry (Icelandic), neither is a superset of the other. This is WHY the intersection outcomes differ.
Vocabulary as FeatureVI items #
The ad-hoc `greekVI` function above implements the Subset Principle
procedurally. Here we define the same vocabulary as `FeatureVI` items
(@cite{halle-marantz-1993}) and prove that `subsetPrinciple` selects
the same exponents. This connects the gender resolution mechanism to
the formal DM vocabulary insertion framework.
Greek vocabulary items as FeatureVI entries (schema 21).
Most specific first: {FEM,MASC} → F, {MASC} → M, {} → N.
Equations
- One or more equations did not get rendered due to their size.
Instances For
BCS vocabulary items as FeatureVI entries (schema 75).
{FEM,ANIM} → F, {INDIV} → M, {} → N.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Subset Principle agrees with ad-hoc greekVI for human mismatch.
Subset Principle agrees with greekVI for inanimate mismatch.
Subset Principle agrees with greekVI for uniform feminine.
Subset Principle agrees with bcsVI for human mismatch.
Same vocabulary + different geometry → different outcome.
Greek and Icelandic share the vocabulary (both use greekVocabItems),
but the Subset Principle yields different results because the geometry
produces different intersections.
FeatureOrder instances #
The `FeatureOrder` structure packages a feature geometry — its nodes
and the bundle function that maps each node to its full set of
entailed features. This formalizes the paper's central insight that
cross-linguistic variation is geometry variation.
Greek feature order: CLASS > MASC > FEM (linear chain).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Icelandic feature order: CLASS > {MASC, FEM} (independent siblings).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Greek entailment: FEM entails MASC (FEM's bundle ⊇ MASC's bundle).
Icelandic: FEM does NOT entail MASC (independent siblings).
Greek: MASC entails CLASS.
Greek: CLASS does NOT entail MASC (entailment is asymmetric).
MRH as a property of geometries #
@cite{adamson-anagnostopoulou-2025}'s Mismatch Resolution Hypothesis:
Greek satisfies MRH because ALL pairwise resolution outcomes produce
non-empty intersections — no default insertion is ever needed.
We verify this via the `satisfiesMRH` predicate from
`GenderResolution`, instantiated with the geometry-derived bundles.
Greek satisfies MRH: all pairwise resolutions succeed.
Icelandic satisfies MRH: all pairwise resolutions succeed. Despite different outcomes (neuter for humans instead of masculine), no resolution yields an empty intersection.
Both satisfy MRH — this is the paper's "no default insertion" claim: the difference between Greek and Icelandic is the content of the intersection, not whether it exists.
Three or more conjuncts #
`resolveN` extends binary resolution to n-ary coordination via
iterated intersection. The predictions are the same: gender-matching
agreement emerges iff all conjuncts share a feature (after percolation).
Greek: three human feminines → {CLASS,MASC,FEM}.
Greek: three mismatched humans → {CLASS,MASC} (FEM eliminated). FEM + MASC + FEM: FEM present in first and third but not second.
Greek: three inanimates → {CLASS} → neuter.
Icelandic: three mismatched humans → {CLASS} → neuter. Because MASC and FEM are independent, any mismatch loses both.
BCS: three mismatched humans → {CLASS,INDIV,MASC,ANIM} → masculine. INDIV guarantees masculine even with mismatched FEM.
N-ary subsumes binary: resolveN [fs1, fs2] = resolve fs1 fs2.