Wang & Sun (2026): Detaching Mandarin Classifiers from Nouns #
@cite{wang-sun-2026} @cite{adger-2025}
Applies @cite{adger-2025}'s mereological syntax to three problems in Mandarin classifier phrases:
- Modification: Degree-modified adjectives cannot appear between Num and Cl — predicted by dimensionality (Q is full).
- Dislocation: [Cl-N] cannot be topicalized independently of Num — predicted by collective spell-out at Q.
- Interpretation: 的 de changes classifier from sortal (concrete object) to mensural (abstract unit) — predicted by cross-dimensional visibility.
Key Structural Insight #
Classifiers are NOT projections of nouns. They are independent syntactic objects that subjoin to Q as its 1-part. The classifier word spells out at Q (collective spell-out of the Q-Cl complementation line). This detachment from N explains all three problems.
Grammaticality status for Mandarin classifier examples.
- ok : GramStatus
- marginal : GramStatus
- bad : GramStatus
- hashBad : GramStatus
Instances For
Equations
- WangSun2026.instDecidableEqGramStatus x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- WangSun2026.instReprGramStatus = { reprPrec := WangSun2026.instReprGramStatus.repr }
- mandarin : String
- gloss : String
- translation : String
- status : GramStatus
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- WangSun2026.instReprClDatum = { reprPrec := WangSun2026.instReprClDatum.repr }
Problem 1: Modification restrictions #
(4c) *yī hěn cōngmíng de gè xuéshēng — degree adj before Cl: bad.
Equations
- WangSun2026.ex4c = { mandarin := "yī hěn cōngmíng de gè xuéshēng", gloss := "one very clever DE CL student", translation := "a very clever student", status := WangSun2026.GramStatus.bad }
Instances For
(8a) yī gè cōngmíng (de) xuéshēng — adj after Cl: ok.
Equations
- WangSun2026.ex8a = { mandarin := "yī gè cōngmíng (de) xuéshēng", gloss := "one CL clever (DE) student", translation := "a clever student", status := WangSun2026.GramStatus.ok }
Instances For
(29a) *yī hěn dà (de) zhāng zhuōzi — degree adj between Num and Cl: bad.
Equations
- WangSun2026.ex29a = { mandarin := "yī hěn dà (de) zhāng zhuōzi", gloss := "one very big (DE) CL table", translation := "a very big table", status := WangSun2026.GramStatus.bad }
Instances For
(29b) yī dà zhāng zhuōzi — bare size adj before Cl: ok (included in Q spell-out).
Equations
- WangSun2026.ex29b = { mandarin := "yī dà zhāng zhuōzi", gloss := "one big CL table", translation := "a big table", status := WangSun2026.GramStatus.ok }
Instances For
Problem 1b: Modification IS possible after Cl #
(28a) yī zhāng hěn dà de zhuōzi — degree adj after Cl: ok.
Equations
- WangSun2026.ex28a = { mandarin := "yī zhāng hěn dà de zhuōzi", gloss := "one CL very big DE table", translation := "a very big table", status := WangSun2026.GramStatus.ok }
Instances For
Problem 2: Dislocation restrictions #
(5c) *Gè píngguǒ, Zhāngsān chī-le sān — [Cl-N] topicalized, Num stranded: bad.
Equations
- One or more equations did not get rendered due to their size.
Instances For
(13b) Píngguǒ, Zhāngsān chī-le sān gè — N topicalized alone: ok.
Equations
- One or more equations did not get rendered due to their size.
Instances For
(14a) *Sān gè, Zhāngsān chī-le píngguǒ — [Num-Cl] without N: bad.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Wh-island effects (§5) #
(39a) Zhāngsān mǎi-le jǐ zhāng hěn dà de zhuōzi? — wh-numeral before Cl: ok.
Equations
- One or more equations did not get rendered due to their size.
Instances For
(39b) *Zhāngsān mǎi-le hěn dà de jǐ zhāng zhuōzi? — wh-numeral after Mod: bad.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Problem 3: Interpretation change with 的 de #
(33a) sān bēi jiǔ — no de: "three glasses of liquor" (real glasses).
Equations
- WangSun2026.ex33a = { mandarin := "sān bēi jiǔ", gloss := "three glass liquor", translation := "three glasses of liquor", status := WangSun2026.GramStatus.ok }
Instances For
(33b) sān bēi de jiǔ — with de: "three glassfuls of liquor" (abstract).
Equations
- WangSun2026.ex33b = { mandarin := "sān bēi de jiǔ", gloss := "three glass DE liquor", translation := "three glassfuls of liquor", status := WangSun2026.GramStatus.ok }
Instances For
(26b): yī zhāng zhuōzi "a table" #
```
D
@Q: zhāng ← Q and Cl collectively spell out
Num: yī Cl
N: zhuōzi
```
Parthood: N <₁ Cl <₁ Q, Num <₂ Q, Q <₁ D.
Instances For
(34a): sān bēi jiǔ "three glasses of liquor" — without 的 #
Same topology as (26b). Cl is in D's 1-part chain → visible → sortal.
Instances For
(34b): sān bēi de jiǔ "three glassfuls of liquor" — with 的 #
的 *de* is the spell-out of Mod. Q subjoins to Mod as 1-part;
Mod then subjoins to D as 2-part. N subjoins directly to D
as 1-part — NOT to Cl (contrast with noDe where N <₁ Cl).
```
D
N: jiǔ Mod: de
@Q: bēi
Cl Num: sān
```
Parthood: Cl <₁ Q, Num <₂ Q, Q <₁ Mod, N <₁ D, Mod <₂ D.
Cl is NOT in D's 1-part chain → invisible → mensural.
Instances For
Instances For
Instances For
(27b): hěn dà de "very big" #
```
Mod
@Deg: dà-de
A Adv: hěn
```
A <₁ Deg, Adv <₂ Deg, Deg <₁ Mod. Mod spells out *de* at its node;
Deg spells out the adjective *dà* collectively with A.
(28c): yī zhāng hěn dà de zhuōzi "a very big table" #
Mod adjoins as 2-part of Cl (Cl has only N as 1-part, so it has room).
```
D
@Q: zhāng
Num: yī Cl
N: zhuōzi Mod
@Deg: dà-de
A Adv: hěn
```
Parthood: A <₁ Deg, Adv <₂ Deg, Deg <₁ Mod, N <₁ Cl, Mod <₂ Cl,
Cl <₁ Q, Num <₂ Q, Q <₁ D.
Instances For
(32b): hěn dà de yī zhāng zhuōzi "a very big table" — pre-DP modifier #
Mod adjoins as 2-part of D (D has only Q as 1-part, so it has room).
Contrast with (28c) where Mod is 2-part of Cl. Same structure
blocks wh-extraction in (39b): D is full, Num cannot subjoin.
```
D
@Q: zhāng Mod: de
Num: yī Cl @Deg: dà-de
N A Adv: hěn
```
Parthood: N <₁ Cl <₁ Q, Num <₂ Q, Q <₁ D, Mod <₂ D.
(45b): sān kē "three (CL)" — measure phrase without N #
Classifier independent of noun: Cl has no N subjoined.
```
@PRED: duō
Deg
Q A
Num: sān Cl: kē
```
Cl <₁ Q, Num <₂ Q, Q <₁ Deg, A <₂ Deg, Deg <₁ Pred.
Instances For
Instances For
Prediction 1: Q is full — no modifier can intervene #
Q has both a 1-part (Cl) and a 2-part (Num). By Dimensionality,
no further subjunction to Q is possible. This derives the ban on
degree-modified adjectives between Num and Cl (examples 4c, 29a).
Cl has only N as 1-part — it has room for Mod as 2-part.
(28c): post-Cl modification preserves Cl visibility from D. Adding Mod as Cl's 2-part does not disrupt the 1-part chain D → Q → Cl, so the classifier retains its sortal reading.
Prediction 1b: Classifier independence from nouns #
(45b) shows a classifier with no N subjoined — the classifier
functions as a measure phrase in a degree construction.
Cl has no N as subpart in (45b): classifier is structurally independent of nouns.
Cl is still in Pred's 1-part chain despite having no N.
Prediction 2: Cl spells out at Q #
Q's complementation line (1-part chain) includes both Q and Cl.
The classifier word is realized at Q via collective spell-out.
Extracting the classifier therefore means extracting Q, which also
contains Num as 2-part — [Cl-N] cannot topicalize without Num
(examples 5c, 14a).
Q contains both Cl and Num: extracting Q extracts both.
Prediction 3: 的 de changes Cl visibility from D #
**Without 的**: Cl <₁ Q <₁ D — within-dimension transitivity applies.
Cl is in D's 1-part chain, so Cl's semantic content (sortal meaning,
e.g., "real glass") contributes to D's referential interpretation.
**With 的**: Cl <₁ Q <₁ Mod ∧₂ D — cross-dimension path. Cl is NOT
in D's 1-part chain (D's 1-part chain goes through N only). Cl's
content is invisible to D → abstract measure reading ("glassful").
The de contrast: same classifier, different visibility.
Prediction 3b: 的 detaches N from Cl #
The key structural difference: without 的, N is 1-part of Cl
(N <₁ Cl); with 的, Cl is bare (no N) and N is directly 1-part
of D (N <₁ D). This is NOT movement — it is a different
derivation path (different subjunction targets).
Without 的: N is inside Cl (N <₁ Cl).
With 的: Cl is bare — no N subjoined to it.
Prediction 3c: Two modifier positions #
Post-Cl (28c): Mod <₂ Cl. D has only Q as 1-part (not full).
Pre-DP (32b): Mod <₂ D. D is full (Q + Mod).
Both are grammatical, but the pre-DP structure blocks
wh-movement because D is full (see Prediction 5).
Post-Cl modifier: D is NOT full (only Q as 1-part).
Prediction 4: N is always visible from D #
Regardless of 的, N is always directly in D's 1-part chain:
without 的, N <₁ Cl <₁ Q <₁ D; with 的, N <₁ D.
The noun's denotation always contributes to the referent.
Prediction 5: Wh-island via dimensionality (§5) #
In (39a), D has only Q as 1-part (no Mod). D is not full, so
Num (the wh-phrase jǐ) can subjoin to D as its 2-part — the
intermediate step needed for wh-movement to C.
In (39b), Mod is 2-part of D (pre-nominal modifier). D is full.
Num cannot subjoin to D → wh-movement is blocked.
(39a): D without Mod has room — Num can subjoin.
(39b): D with Mod is full — Num blocked.
The wh-island contrast: same Q, different D dimensionality.
Reading of a classifier determined by its structural visibility from D. When Cl is in D's 1-part chain, the classifier denotes a concrete object (sortal). When invisible, it denotes an abstract unit (mensural).
Instances For
Equations
- WangSun2026.instDecidableEqClReading x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- WangSun2026.instReprClReading = { reprPrec := WangSun2026.instReprClReading.repr }
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Both modifier positions (post-Cl and pre-DP) preserve Cl visibility from D, so the classifier retains its sortal reading regardless of where Mod attaches.
杯 bēi (glass) is the classifier in examples (33)–(34). It has
isMensural = true, indicating it CAN function as a measure phrase.
Whether it DOES have a mensural reading is determined by the structure
(§4 above), not by the lexical entry alone.
张 zhāng (flat surface) is the classifier in the basic structure (26b) and modification examples (28)–(29). It is a pure sortal classifier — not lexically mensural.
个 gè (general) is the default classifier in the dislocation examples (5c), (13b), (14a).
Both @cite{borer-2005} and @cite{wang-sun-2026} place Q below D and above N in the nominal spine, and both treat classifiers as independent of nouns. The mereological analysis adds a structural explanation for the 的-contrast that Borer's theory does not directly address: without 的, the classifier is visible to D (sortal); with 的, it is invisible (mensural).
The 1-part chain from D (without 的) follows the ordering D → Q → Cl → N. Two differences from @cite{borer-2005}'s EP (D → Num → Q → n → N): (1) n (categorizer) is absent in the mereological analysis; (2) Num is Q's 2-part here (outside the 1-part chain), whereas Borer places Num above Q in the spine. Both frameworks agree on D > Q > N ordering and on Q as the individuation locus.
Q is the individuation locus in both frameworks. Borer: Q hosts CL#/Div, converting CUM → QUA. Wang & Sun: Cl subjoins to Q as 1-part; Q spells out the classifier.
End-to-end argumentation for 杯 bēi:
- Fragment: bēi has mensural affordance (
isMensural = true) - Without 的: Cl visible from D → sortal reading (real glass)
- With 的: Cl invisible from D → mensural reading (glassful)
- Without-的 spine matches @cite{borer-2005}'s nominal EP
End-to-end for modification: Cl has room for Mod (derives 28c), Q is full (blocks 29a), and post-Cl modification preserves Cl visibility from D (sortal reading maintained).
The structural predictions in §§ 3–5 determine syntactic visibility.
This section connects visibility to @cite{borer-2005}'s mereological
semantics via the bridge in Interpretation.lean: visible Cl → QUA
(count/sortal); invisible Cl → root preserved (mass/mensural).
Without 的: Cl visible → denotation is individuated → QUA.
With 的: Cl invisible → denotation equals the root predicate.
With 的 and a cumulative root: denotation is CUM (mass/mensural).
End-to-end 的-contrast: same root, different structure, opposite mereological properties.
End-to-end argumentation connecting three independent theories:
1. **@cite{chierchia-1998}**: Mandarin is [+arg, -pred] → nouns are
kind-denoting → bare nouns are arguments → classifiers required.
2. **@cite{borer-2005}**: Roots denote cumulative (CUM) predicates.
The Q head hosts individuation (Div), converting CUM → QUA.
Classifiers spell out at Q.
3. **@cite{wang-sun-2026}**: Mereological syntax determines whether
Cl is visible from D via 1-part chain transitivity. Visible Cl → QUA
(sortal); invisible Cl → CUM (mensural). The particle 的 changes
dimensional attachment, toggling visibility.
The CUM P hypothesis in `bei_semantic_contrast` IS Borer's thesis
instantiated for Chierchia's [+arg, -pred] languages: root predicates
are cumulative because they denote kind-level extensions (closed under
join).
Mandarin's NMP is [+arg, -pred]: nouns denote kinds, requiring classifiers for counting. This is the typological precondition for the entire mereological analysis.
The Chierchia–Borer–Wang&Sun chain for 杯 bēi:
- Mandarin NMP = argOnly → classifiers required (Chierchia)
- bēi has mensural affordance (Fragment)
- Without 的: Cl visible → QUA (Borer + Wang&Sun)
- With 的: Cl invisible → CUM under Borer's thesis
- Same root, opposite mereological properties (Interpretation)