2.1. One surface template, multiple semantic input types

The surface configuration XP-de-NP is pervasive in Mandarin. The modifier XP can be nominal, adjectival or clausal, yielding a correspondingly wide range of XP–de–NP interpretations.

While examples (1)–(5) typically motivate ‘many-de’ analyses, we take the opposite approach. Treating de as uniform, we ask what the XP’s semantic type and the head noun’s lexical resources must satisfy to derive these interpretations compositionally.

2.2. Three clausal subtypes under de: NCCs, SECs and GRCs

We identify three major types of prenominal clausal modifiers followed by de and adopt the following abbreviations:

1. A. NCC (noun complement clause): a clause typically interpreted as the propositional content of nouns like yaoyan ‘rumor’ or xiaoxi ‘news’.

2. B. SEC (saturated event clause or gapless relative): a clause with an eventive profile, typically combining with event-correlate nouns such as shengyin ‘sound’ or weidao ‘smell’.

3. C. GRC (gapped relative clause or restrictive relative): a clause containing a gap and interpreted as a property of individuals.

These three clausal subtypes exhibit distinct behaviors under stacking. Section 6 revisits these contrasts and derives the corresponding ordering effects.

2.3. Predicate-oriented AP-de modifiers

Finally, Mandarin permits non‑predicative, non‑restrictive and intensional AP modifiers such as weiyi ‘only/unique’ and qianren ‘former’ in AP–de–NP constructions (see also Paul Reference Paul2005).

Unlike previously discussed APs, these modifiers are non‑predicative and non‑restrictive, and they often encode presuppositional or temporal content (e.g. uniqueness, temporal precedence). In our unified analysis developed below, this contrast is accounted for not by positing a separate lexical entry for de, but by the AP modifier’s own semantic type and lexical content.

2.4. Eventive fake nominal modification

A distinct puzzle concerns cases where a DP-de-NP subject combines with an event-evaluative predicate, yielding an eventive interpretation. This phenomenon is central to the ‘fake modification’ analyses in Huang (Reference Huang2008), Huang, Li & Li (Reference Huang, Li and Li2009), Wenshan Li (Reference Li2021) and Zhang & Liu (Reference Zhang, Liu, Dong, Hong, Lin and Jin2024), though their approaches differ substantially. In these constructions, the syntax appears to involve DP‑internal modification, yet the semantics align with event predication shaped by construction‑specific dependencies, creating a syntax-semantics mismatch (Huang, Li & Li Reference Huang, Li and Li2009). Representative examples are provided in (7a, b).

The subject phrase Daiziying de yuqiu in (7a) is not naturally interpreted as a possessed shuttlecock object; rather, it denotes an eventive description, meaning Daiziying’s playing badminton. This eventive interpretation is supported by the verb-copying paraphrase in (8), which makes the underlying event predication overt.Footnote ³

In the durative subtype in (7b), the measure phrase does not quantify the nominal individual (song); instead, it imposes a duration construal on the associated singing eventuality, acting as an ‘event delimiter’ (Zhang & Liu Reference Zhang, Liu, Dong, Hong, Lin and Jin2024). Later, we argue that the two eventive patterns follow distinct compositional routes.

3.1. Descriptive and functional-typological perspectives

Early structuralist and functional-typological accounts focused on distributional classifications of de (attributive, adverbial or nominalizing; e.g. Zhu Reference Zhu1961, Reference Zhu1966), or its role as a general prenominal modification marker capable of packaging words, phrases and clauses as nominal modifiers (e.g. Li & Thompson Reference Li and Thompson1981, Guoying Huang Reference Huang1982, Huang & Shi Reference Huang and Shi2016). While these accounts establish de’s descriptive range, they leave unresolved how the XP-NP relation is compositionally derived and constrained.

3.2. Formal syntactic analyses

While structuralist and functional studies provide distributional classifications, formal syntactic approaches investigate what kind of head de is and where it merges. Given its occurrence with adjectival, nominal and clausal prenominal modifiers, a recurring hypothesis is that surface uniformity conceals structural heterogeneity. Depending on the configuration, scholars have instantiated de as distinct functional material, such as a complementizer, determiner, relator, conjunction-like element or relativizer (see Cheng Reference Cheng1986, Simpson Reference Simpson2001, Reference Simpson, Tang and Liu2002, den Dikken & Singhapreecha Reference den Dikken and Singhapreecha2004, Paul Reference Paul2005, den Dikken Reference den Dikken2006, Li Reference Li2008, Paul Reference Paul2012).

Consequently, syntactic approaches conclude that not all de-marked modifiers are structurally identical. Clausal de‑modification, in particular, has been argued to diverge into distinct subtypes (Huang Reference Huang2016, among others), with diagnostics separating relative-like, complement-like and other prenominal clausal patterns. We return to this issue below, distinguishing propositional/content‑like modification (NCC‑like), eventive prenominal clauses (SEC‑like) and gapped relatives (GRC‑like).

Purely structural analyses, however, leave the central semantic problem unresolved: they do not determine the relation linking XP and NP or what constrains these possibilities. A successful account must specify both the semantic type contributed by XP and the mechanism introducing and constraining the XP-NP relation.

3.3. Semantic analysis

Semantic approaches shift focus from de’s syntactic category to its compositional role. Since an XP can denote an individual, measure value, property or proposition, and the head noun licenses various relations to it, a semantic analysis must specify the XP’s type, de’s semantic contribution and constraints on the resulting XP-NP relation.

Within this domain, two influential strands are Huang’s (Reference Huang2006) property-theoretic program and Lu & Pan’s (Reference Lu and Pan2016) and Pan & Lu’s (Reference Pan and Lu2021) type-based linker framework. While both treat de as an active ingredient in semantic composition rather than a mere syntactic marker, their explanatory foci differ. Huang emphasizes type-mismatch repair via nominalization or type shifting, whereas Lu & Pan distinguish two types of linking: predicate-type modification versus argumental dependency.

3.4. From linking to constrained relation resolution

When a theory permits de to link argument-like or clausal modifiers to a head noun (as in Lu & Pan’s de_a component), the key remaining task is no longer establishing whether linking is possible, but rather determining how the link is interpreted and constrained.

A natural framework comes from the semantics of genitives and relational nouns, which treat genitive-like morphology as introducing an underspecified relation variable R, whose value is fixed or sharply restricted by lexical and contextual information (Partee Reference Partee, van Benthem and ter Meulen1983/1997, Partee & Borschev Reference Partee, Borschev, Riehl and Daly2000, Vikner & Jensen Reference Vikner and Jensen2002, Partee & Borschev Reference Partee, Borschev, Lang, Maienborn and Fabricius-Hansen2003). The methodological lesson for Mandarin de is straightforward: positing a relation variable requires principled constraints on admissible R-values to prevent interpretive arbitrariness.

Generative Lexicon (GL) theory supplies exactly the sort of lexical constraint such a framework requires, as it structures noun meaning internally via Qualia Structure and provides controlled mechanisms for accessing that structure in composition (Pustejovsky Reference Pustejovsky and Pustejovsky1993, Reference Pustejovsky1995, Pustejovsky & Ježek Reference Pustejovsky and Ježek2008). Importantly for the metalanguage of this paper, GL distinguishes routine qualia-guided linking (selective binding) from genuine type-mismatch repair (coercion). Keeping this distinction explicit is essential because much of de’s relational variability resembles lexically guided role targeting, while only a limited subset of cases plausibly involves repair driven by selectional pressure.

Similarly, Mandarin research on wùxìng/qualia roles likewise argues that implicit-relational readings inside nominal modification are lexically constrained (Wei & Yuan Reference Wei and Yuan2013, Yuan Reference Yuan2014, Li Qiang Reference Li2015, Li Xianhe Reference Li2021). However, this literature remains largely descriptive and typically lacks the implementation of a single-denotation linker with an explicit R-variable regulated by a head-driven qualia-based constraint.

Accordingly, Section 4 implements a unified architecture that (i) treats de as a single semantic linker introducing an underspecified R and (ii) restricts admissible values of R to relations supplied by the lexical semantics of the nominal head or a small inventory of grammar-given interface relations. Context then selects among the remaining compatible candidates only when the grammar underdetermines the choice.

4.1. GL background: qualia structure and distinct generative devices

Within Pustejovsky’s (Reference Pustejovsky1995) Generative Lexicon framework, the qualia structure for a lexical item offers a structured representation of word meaning beyond simple sense listing, enabling dynamic composition, polysemy resolution and type coercion. For nouns, this structure captures their essential semantic dimensions through a limited set of qualia roles functioning as compositional targets.

(9)	Qualia Structure of Nouns
	(a)	formal: what kind of entity the noun denotes and what intrinsic properties or dimensions distinguish it within a larger domain.
		$ {\mathrm{Q}}_{\mathrm{FORMAL}}(N)=\unicode{x03BB} u\;\unicode{x03BB} x\left[{\mathrm{FORMAL}}_N\left(u,x\right)\right] $
		(u may be a classifying property of x or a value on an intrinsic dimension.)
	(b)	constitutive: what the denoted entity is made of, what parts or components it has or what constitutes its content.
		$ {\mathrm{Q}}_{\mathrm{CONSTITUTIVE}}(N)=\unicode{x03BB} u\;\unicode{x03BB} x\left[{\mathrm{CONSTITUTIVE}}_N\left(u,x\right)\right] $
		(u is typically a material/part but can be a measure value.)
	(c)	telic: the purpose, function, characteristic use or characteristic activity associated with the denoted entity.
		$ {\mathrm{Q}}_{\mathrm{TELIC}}(N)=\lambda {u}_e\lambda {x}_e\lambda {e}_v\left[{\mathrm{TELIC}}_N\left(e,u,x\right)\right] $
		(e is a TELIC-associated event involving the nominal individual x and the role filler u; $ v $ is the type of event.)
	(d)	agentive: how the denoted entity comes into being, including its source, creator or characteristic mode of production.
		$ {\mathrm{Q}}_{\mathrm{AGENTIVE}}(N)=\lambda {u}_e\lambda {x}_e\lambda {e}_v\left[{\mathrm{AGENTIVE}}_N\left(e,u,x\right)\right] $
		(e is an AGENTIVE-associated event through which x originates, with u as the relevant source or creator.)

These descriptions characterize the lexical content of the four qualia roles without imposing a uniform argument structure on them. A non-event-dependent role may relate a role filler u directly to the nominal individual x, whereas an event-dependent role may additionally introduce an event argument e. Footnote ⁴ The corresponding relation schemas are made explicit by the Selective Binding rules in (10). Compositional processes then determine whether any argument is subsequently saturated or existentially closed.

Furthermore, GL distinguishes two mechanisms for disambiguating lexical meaning in composition. Selective binding applies when a modifier composes with a noun by targeting a specific internal role, typically a role-instantiated relation, rather than directly modifying the entity denoted by the noun. Classic cases include adjective–noun combinations such as fast car and fast typist, where fast predicates over a qualia-licensed event associated with the noun (e.g. driving/typing) rather than the physical object or person. Accordingly, selective binding can yield event-dependent predicates when it targets the TELIC or AGENTIVE quale.

Coercion, by contrast, is a repair mechanism triggered by type mismatch. It applies when a selecting expression requires a semantic type that its argument does not provide, and no facet of the argument directly satisfies the predicate’s requirements. A canonical illustration is They want a 24-hour library, where want selects a proposition but library denotes an entity. In this case, coercion shifts the NP into a propositional reading, yielding the interpretation ‘They want [there to be] a 24-hour library’.

4.2. Relation variables and genitives: motivation for R and lexical constraint

The interpretive flexibility of Mandarin DP-de-NP mirrors the well‑documented variability of English genitives (DP’s NP). A standard semantic approach to genitives introduces a free relation variable R (Partee Reference Partee, van Benthem and ter Meulen1983/1997, Partee & Borschev Reference Partee, Borschev, Cooper and Gamkrelidze1998), whose value is determined by either lexical information (especially with relational nouns) or contextual resolution (for non‑relational nouns).

Building on this, GL-oriented analyses (e.g. Vikner & Jensen Reference Vikner and Jensen2002) further stress that the head noun’s lexical semantics, particularly its qualia structure, limits which relations count as natural, thereby reducing reliance on unconstrained pragmatics. We draw the same conclusion for Mandarin de, but implement it via an explicit head-driven filter and a strict division of labor between selective binding and coercion.

4.2.1. Selective binding as a qualia-relation generator

Pustejovsky (Reference Pustejovsky1995, esp. Section 7.3) introduces ‘selective binding’ (hereafter SB) to demonstrate that modification need not apply to the noun’s entire referent; rather, it can target an internal qualia role encoded in the noun’s lexical meaning (cf. fast typist). Building on this concept and Vikner & Jensen’s (Reference Vikner and Jensen2002) genitive implementation, we treat selective binding as a general rule for retrieving and instantiating a qualia role as a role-instantiated relation.

If a noun’s argument structure is not inherently relational, selective binding targets a qualia role q contributed by the noun, yielding a role-instantiated relation that serves as a candidate value for the variable R. Formula (10a) represents the non-event-dependent case,Footnote ⁵ while (10b) illustrates the event-dependent case. If the targeted qualia relation includes an event argument that remains open for further composition, selective binding generates an event-dependent relation.

Furthermore, within this framework, selective binding targets a nominal constituent Nom headed by N (abbreviated as N-indexed Nom_N) rather than the bare N alone. This accounts for the fact that XP-de modifiers can stack, allowing an unsaturated qualia role to be inherited across a larger nominal constituent.

(10)	selective binding
	(a)	$ {\mathrm{SB}}_q\left({Nom}_N\right)=\unicode{x03BB} {u}_{\tau}\unicode{x03BB} {x}_e.[ Nom^{\prime }(x)\wedge {Q}_q(N)(u)(x)] $ type: $ \left\langle \tau, \left\langle e,t\right\rangle \right\rangle $ .
	(b)	$ {\mathrm{SB}}_q\left({Nom}_N\right)=\lambda {u}_{\tau}\lambda {x}_e\lambda {e}_v[ Nom^{\prime }(x)\wedge {Q}_q(N)(u)(x)(e)] $ type: $ \left\langle \tau, \left\langle e,\left\langle v,t\right\rangle \right\rangle \right\rangle $ .

For each role q, Q_q(N) represents the noun-specific relational content associated with that role. Accordingly, the schemas in (10) distinguish whether the relation derived by SB is non-event-dependent or whether it preserves an event argument for subsequent composition. Specifically, in (10a), SB(Nom_N) links an individual x satisfying the description of Nom to a role filler u that occupies N’s q-role with respect to x. In (10b), the same relation additionally introduces an open event argument e, which remains available for further composition. For each role q, Q_q(N) returns the relational content specific to N associated with that role. The type τ is determined by the modifier-side argument licensed by the targeted relation: it can be e, v, m, ⟨e,t⟩, ⟨s,t⟩, ⟨⟨v,t⟩,⟨v,t⟩⟩ and so forth.

Selective binding relations often introduce implicit, unidentified arguments lacking independent reference. To ensure type-driven composition proceeds smoothly, we propose that these arguments undergo existential closure at the appropriate structural boundaries (see also Pustejovsky Reference Pustejovsky1995). Specifically, we assume two modes of existential closure: outermost existential closure (cf. Heim Reference Heim1982, Diesing Reference Diesing1992) and event closure for structurally internal event arguments.

(11)	existential closure (EC)
	(a)	Outermost EC (ECouter)
		$ {\mathrm{EC}}_{\mathrm{outer}}\left(\lambda y.\lambda {x}_1\dots \lambda {x}_n.P\left(y,{x}_1,\dots, {x}_n\right)\right)=\lambda {x}_1\dots \lambda {x}_n.\exists y\left[P\left(y,{x}_1,\dots, {x}_n\right)\right] $
	(b)	Event Closure (ECevent)
		$ {\mathrm{EC}}_{\mathrm{event}}\left(\lambda {x}_1\dots \lambda {x}_n{\lambda e}_v\left[R\left({x}_1,\dots {x}_n,e\right)\right]\right)=\lambda {x}_1\dots \lambda {x}_n.\exists {e}_v\left[R\left({x}_1,\dots {x}_n,e\right)\right] $

EC applies only when closure produces the type required for subsequent composition.

4.3. The semantics of de

4.3.2. de as a polymorphic linker that introduces R

In line with Partee’s treatment of the genitive ’s, we analyze de as a semantic linker. It first combines with the meaning of the modifier XP to introduce an underspecified generalized relation R, ultimately yielding a function over the individual argument associated with the post-de nominal, as illustrated in (13).

Specifically, the variables in (13) are defined as follows: α_τ represents the denotation of the pre-de modifier XP; R_{⟨τ,⟨e,ρ⟩⟩} serves as a linking-relation variable of type ⟨τ,⟨e,ρ⟩⟩; x _e is the individual argument of the post-de nominal; and ρ denotes the output type.

(13)	Core linker meaning
	$ [[\mathrm{d}\mathrm{e}]]=\lambda {\alpha}_{\tau}\lambda {R}_{\langle \tau, \langle e,\rho \rangle \rangle}\lambda {x}_e.R(\alpha )(x) $

As noted previously, the admissible values of R stem from relations supplied by the noun’s lexical semantics, including inherent and qualia‑instantiated relations, together with grammar‑driven interface relations. We formalize this as the following admissibility constraint:

(14)	Admissible values of R
	(a)	$ R\in \mathrm{SourceR}(N) $
	(b)	$ \mathrm{SourceR}(N)=\mathrm{SourceR}\left({N}_{\mathrm{Rel}}\right)\cup \mathrm{SourceR}\left({N}_{\mathrm{Qualia}}\right)\cup \mathrm{SourceR}\left({N}_{\mathrm{Intf}}\right) $

When multiple members of SourceR(N) are type‑ and sort‑compatible with both the modifier and the nominal constituent, context determines the appropriate candidate.

Once the linker is supplied with the selected relation R, it yields the output in (15).

(15)	Linker output after relation resolution
	$ [\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ]([\hskip0.1em [\mathrm{X}\mathrm{P}]\hskip0.1em ])(R)=\lambda {x}_e.{R}_{\langle \tau, \langle e,\rho \rangle \rangle }([\hskip0.1em [\mathrm{X}\mathrm{P}]\hskip0.1em ])(x) $

This operation establishes a relation between the XP-de constituent and the individual x without explicitly associating x with its Nom constituent (namely, the NP in an XP-de-NP construction). Yet, this dissociation is merely apparent, since restriction on x originates from the relation R. For instance, when R is a transitive relational noun such as fuqin ‘father’, x functions as a pure argument of the relational noun; conversely, when R is a qualia relation derived via selective binding, the restriction on x is entailed by the Nom incorporated into the SB rule (see (10)).

What merits closer attention is the output type of R _{⟨τ,⟨e,ρ⟩} ( $ [\hskip0.1em [\mathrm{X}\mathrm{P}]\hskip0.1em ] $ )(x). Since the first argument $ [\hskip0.1em [\mathrm{X}\mathrm{P}]\hskip0.1em ] $ is of type τ and x is of type e, the output type of R _{⟨τ,⟨e,ρ⟩} ( $ [\hskip0.1em [\mathrm{X}\mathrm{P}]\hskip0.1em ] $ )(x) must be of type ρ. Within this framework, ρ can be either type t or ⟨v,t⟩, depending on the targeted qualia-derived relation and its event-dependency. Specifically, if the output type is ⟨v, t⟩, the event variable remains in the output; conversely, if the output type is ⟨e,t⟩, it functions as a standard individual property.

5.1. DP-de-NP: agentive and possessive/control readings

The phrase Zhangsan de shu ‘Zhangsan’s book’ is ambiguous between ‘the book written by Zhangsan’ and ‘the book Zhangsan owns’. Under our proposed framework, this ambiguity arises because the noun shu ‘book’ supplies multiple available relations, allowing context to select among them. Consequently, this example illustrates the core use case of the relation-variable architecture: the semantic function of de remains constant, while the relation variable R is selected from the head noun’s set of admissible relations. For the case under discussion, de is instantiated as shown in (16):

(16)	de in DP-de-NP structure
	$ [[\mathrm{d}\mathrm{e}]]=\lambda {x}_e.\lambda {R}_{\langle e,\langle e,t\rangle \rangle }.\lambda {y}_e.R(x)(y) $

In ordinary DP-de-NP interpretations, an event-dependent qualia relation generated by SB (10b) may undergo event closure EC_event on its event argument, yielding a two-place relation of type ⟨e,⟨e,t⟩⟩.

5.1.1. Agentive (authorship) reading

The authorship reading derives from the head noun’s AGENTIVE role, which is accessed via selective binding. Since the creation event argument is not required for subsequent composition, it is existentially closed by existential closure (EC) as defined in (11).

(17)	Lexical entry of shu ‘book’
	(a)	Argument structure: $ [[\mathrm{s}\mathrm{h}\mathrm{u}]]=\lambda {x}_e.\mathrm{b}\mathrm{o}\mathrm{o}\mathrm{k}(x) $
	(b)	Agentive relation for shu
		$ {\mathrm{SB}}_{\mathrm{AGENTIVE}}\left(\mathrm{shu}\right)=\lambda y\lambda x\lambda v\left[\begin{array}{c}\mathrm{book}(x)\wedge \mathrm{write}(v)\\ {}\wedge \mathrm{Agent}\left(v,y\right)\wedge \mathrm{Product}\left(v,x\right)\end{array}\right] $
	(c)	$ {\mathrm{EC}}_{\mathrm{event}}\left({\mathrm{SB}}_{\mathrm{AGENTIVE}}\left(\mathrm{shu}\right)\right)=\lambda y\lambda x\exists v\left[\begin{array}{c}\mathrm{book}(x)\wedge \mathrm{write}(v)\\ {}\wedge \mathrm{Agent}\left(v,y\right)\wedge \mathrm{Product}\left(v,x\right)\end{array}\right] $

Following this, the noun phrase Zhangsan de shu ‘Zhangsan’s book’ is computed as follows, yielding an individual property. In this derivation, because the variable x’s restriction book(x) is built into the definition of SB_AGENTIVE(shu), the variable x is automatically restricted to book(x).

(18)	(a)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}]\hskip0.1em ]=\mathrm{Z}\mathrm{S} $
	(b)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.35em \mathrm{d}\mathrm{e}]\hskip0.35em ]=[\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ](\mathrm{Z}\mathrm{S})=\lambda R\lambda x.R(\mathrm{Z}\mathrm{S})(x) $
	(c)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.35em \mathrm{d}\mathrm{e}\hskip0.35em \mathrm{s}\mathrm{h}\mathrm{u}]\hskip0.1em ]=(\lambda R\lambda x.R(\mathrm{Z}\mathrm{S})(x))({\mathrm{EC}}_{\mathrm{event}}({\mathrm{SB}}_{\mathrm{AGENTIVE}}(\mathrm{s}\mathrm{h}\mathrm{u}))) $
		$ =\lambda {x}_{\mathrm{e}}\mathrm{\exists}{v}_{\mathrm{v}}.[\mathrm{b}\mathrm{o}\mathrm{o}\mathrm{k}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{w}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}(v)\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(v,\mathrm{Z}\mathrm{S})\hskip0.35em \wedge \hskip0.35em \mathrm{P}\mathrm{r}\mathrm{o}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{t}(v,x)] $

5.1.2. Possessive/control reading

The possessive reading is derived similarly, but with the variable R resolved to a different lexically licensed relation. Following Vikner & Jensen (Reference Vikner and Jensen2002: 196), we treat possession as a relation broadly available for artifacts. Consequently, for a noun like book, beyond its basic TELIC role (‘to be read’), we assume it may support a facet of possession, control or use (i.e. having the object under one’s authority or effective use). We model this as another admissible relation:

(19)	Possession relation
	$ {R}_{\mathrm{POSSESS}}={\mathrm{SB}}_{\mathrm{POSSESS}}(\mathrm{s}\mathrm{h}\mathrm{u})=\lambda {y}_e\hskip2pt \lambda {x}_e.[\mathrm{b}\mathrm{o}\mathrm{o}\mathrm{k}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{s}\mathrm{s}(y,x)] $

(20)	Possession composition
	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip2pt \mathrm{d}\mathrm{e}\hskip2pt \mathrm{s}\mathrm{h}\mathrm{u}]\hskip0.1em ] $
	$ =[\lambda R\lambda x.R(\mathrm{Z}\mathrm{S})(x)]({\mathrm{SB}}_{\mathrm{POSSESS}}(\mathrm{s}\mathrm{h}\mathrm{u})) $
	$ =\lambda x.\mathrm{b}\mathrm{o}\mathrm{o}\mathrm{k}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{s}\mathrm{s}(\mathrm{Z}\mathrm{S},x) $

Accordingly, both readings of Zhangsan de shu are derived by selecting different head-licensed relations from the lexical resources of shu, while the semantic contribution of de remains constant.

5.2. Num-CL-de-NP: modifier readings and the exclusion of counting

Many scholars have observed that Mandarin Num‑CL‑N sequences permit both counting/quantificational (‘n individuals’) and measure/modifier readings (‘an amount or unit characterized by Num‑CL’). Post‑classifier de strongly favors the latter, yielding an attributive, measure‑oriented interpretation rather than ordinary counting (Tang Reference Tang1990, Reference Tang1993, Reference Tang2005, Xu‑Ping Li Reference Li2011, Li & Rothstein Reference Li and Rothstein2012, among others).

We argue that this contrast arises from the interaction between the typed linker and the head‑licensed admissibility condition. Specifically, a Num-CL sequence takes de to form an XP-de-NP structure only when acting as a non-counting modifier expressing an admissible noun-licensed relation. Demonstrating this, Sections 5.2.1 and 5.2.2 apply these constraints to intrinsic-dimension and amount/unit measurements, respectively. Section 5.2.3 then explains why ordinary counting construals fail to license de.

5.2.1. Intrinsic-dimension measurement

Consider san‑bang de xigua ‘a three‑pound watermelon’. The measure phrase san‑bang (‘3 lb’) specifies the watermelon’s intrinsic weight, distinguishing it from, for instance, the two‑ or four‑pound subclasses. This intrinsic-dimension measurement aligns with the FORMAL role, situating objects within a broader classificatory domain.

(21)	Formal-licensed intrinsic-dimension measurement relation
	(a)	$ [\hskip0.1em [\mathrm{x}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}]\hskip0.1em ]=\lambda x.\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{m}\mathrm{e}\mathrm{l}\mathrm{o}\mathrm{n}(x) $
	(b)	$ {\mathrm{SB}}_{\mathrm{FORMAL}}(\mathrm{x}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a})\in \mathrm{S}\mathrm{o}\mathrm{u}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{R}(\mathrm{x}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}) $
		$ {\mathrm{SB}}_{\mathrm{FORMAL}}\left(\mathrm{xigua}\right)=\lambda m\lambda x.\left[\mathrm{watermelon}(x)\wedge \mathrm{weight}(x)=m\right] $

(22)	(a)	$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}-\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{g}]\hskip0.1em ]=3\hskip2pt \mathrm{l}\mathrm{b} $
	(b)	$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}-\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{g}\hskip2pt \mathrm{d}\mathrm{e}]\hskip0.1em ]=[\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ]([\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}-\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{g}]\hskip0.1em ])=\lambda R\lambda x.R(3\hskip2pt \mathrm{l}\mathrm{b})(x) $
	(c)	$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}-\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{g}\hskip2pt \mathrm{d}\mathrm{e}\hskip2pt \mathrm{x}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}]\hskip0.1em ] $
		$ =\lambda x.\left[\mathrm{watermelon}(x)\wedge \mathrm{weight}(x)=3\;\mathrm{lb}\right] $

5.2.2. Amount/unit measurement

Consider this example from Yen-Hui Audrey Li (Reference Li2013: 105):

(23)	Wo	yi-kouqi	chi-wan	yi-da-ge	de	xigua,	zhang-si-le.
	I	one-breath	eat-finish	one-big-clf	de	watermelon	full-dead-sfp
	‘I finished eating a big watermelon (in amount) in one breath; I am too full.’

Here, de is optional (Audrey Li Reference Li2013: 99, 107), but it typically requires perceptual or classificatory material like da ‘big’; without this adjectival enrichment, de insertion often degrades. Unlike (22), the Num‑Adj‑Cl‑de sequence in (23) measures the actual amount or unit of watermelon consumed rather than classifying it within a measurable subclass. While the modifier yi-da-ge remains a measure-denoting expression of type m, it quantifies the watermelon’s material or mereological constitution.

Thus, the relevant qualia role is CONSTITUTIVE. Recall from (9b) that the relation CONSTITUTIVE_N(u)(x) normally describes an object $ x $ ’s material or parts u. However, because a physical object is also constituted by its spatial extent and physical mass, an intrinsic measurement (e.g. length, weight or volume) specifies the amount of matter or space constituting it. Here, the argument u is of type measure value (or degree/amount), as shown below:

(24)	$ {\mathrm{CONSTITUTIVE}}_N(m)(x)\Longleftrightarrow N(x)\wedge \mathrm{Unit}\_\mathrm{Measure}(x)=m $

Consequently, the CONSTITUTIVE-induced relation for example (23) is shown in (25) with its semantic composition given in (26):

(25)	CONSTITUTIVE-licensed amount/unit-measurement relation
	(a)	$ [\hskip0.1em [\mathrm{x}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}]\hskip0.1em ]=\lambda x.\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{m}\mathrm{e}\mathrm{l}\mathrm{o}\mathrm{n}(x) $
	(b)	$ {\mathrm{SB}}_{\mathrm{CONSTITUTIVE}}\left(\mathrm{xigua}\right)\in \mathrm{SourceR}\left(\mathrm{xigua}\right) $
		$ {\mathrm{SB}}_{\mathrm{CONSTITUTIVE}}\left(\mathrm{xigua}\right) $
		$ =\lambda m\lambda x.\left[\mathrm{watermelon}(x)\wedge \mathrm{Unit}\_\mathrm{Measure}(x)=m\right] $

(26)	(a)	$ [\hskip0.1em [\mathrm{y}\mathrm{i}-\mathrm{d}\mathrm{a}-\mathrm{g}\mathrm{e}]\hskip0.1em ]=\mathrm{o}\mathrm{n}\mathrm{e}\_\mathrm{b}\mathrm{i}\mathrm{g}\_\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}\hskip2pt (\mathrm{t}\mathrm{y}\mathrm{p}\mathrm{e}\hskip2pt m) $
	(b)	$ [\hskip0.1em [\mathrm{y}\mathrm{i}-\mathrm{d}\mathrm{a}-\mathrm{g}\mathrm{e}\hskip2pt \mathrm{d}\mathrm{e}]\hskip0.1em ]=[\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ](\mathrm{o}\mathrm{n}\mathrm{e}\_\mathrm{b}\mathrm{i}\mathrm{g}\_\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}) $
		$ =\lambda R\lambda x.R\left(\mathrm{one}\_\mathrm{big}\_\mathrm{unit}\right)(x) $
	(c)	$ [\hskip0.1em [\mathrm{y}\mathrm{i}-\mathrm{d}\mathrm{a}-\mathrm{g}\mathrm{e}\hskip2pt \mathrm{d}\mathrm{e}\hskip2pt \mathrm{x}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}]\hskip0.1em ] $
		$ =\lambda x.\left[\mathrm{watermelon}(x)\wedge \mathrm{Unit}\_\mathrm{Measure}(x)=\mathrm{one}\_\mathrm{big}\_\mathrm{unit}\right] $

This yields the intended quantity reading: ‘watermelon measured as one big unit’ (or ‘a big unit/amount of watermelon’). Thus, (21) and (25) instantiate the same measure-linking type, differing only in which head-licensed measurement relation supplies the value of R.

5.3. Sequential composition and ordering of multiple DP-de modifiers

The framework naturally accommodates complex nominals with multiple de-modifiers. Each modification layer introduces a relation variable R resolved via the head noun’s lexical semantics. These modifiers operate via sequential selective binding, each saturating a distinct relation within the head noun’s qualia structure. The resulting nominal remains semantically coherent only if necessary lexical resources remain accessible across all derivational layers.

5.3.1. Composition of multiple DP-de modifiers

Consider the following grammatical example (adapted from Lu & Pan Reference Lu and Pan2019: 48):

(28)	Zhangsan	de	beijiguang	de	zhaopian
	Zhangsan	de	aurora	de	photo
	‘A photo of the aurora taken by Zhangsan’

The head noun zhaopian ‘photo’ licenses multiple semantic relations with its modifiers: a content/theme relation (specifying what the photo depicts) and an agent/creator relation (specifying who took it). Lu and Pan observe that, empirically, the content modifier occurs closer to the head noun than the agent modifier. To account for how these lexical relations remain accessible across successive layers of de-modification, we adopt the following inheritance constraint:

1. (29) qualia preservation principle (QPP)

   When a nominal projection headed by a noun N is built up through successive modification, N’s qualia structure remains accessible at every level of the nominal projection. If an expression saturates a specific relation anchored in N’s qualia structure, the resulting projection inherits that structure with the targeted relation now marked as bound (i.e. used). Subsequent modification can then target only remaining unbound qualia-licensed relations.

The QPP formalizes two interface commitments: (i) preserving the lexical head’s qualia inventory across recursive nominal expansion, and (ii) resource sensitivity, preventing a previously saturated qualia-anchored relation from being saturated again.

While the QPP builds on Pustejovsky’s treatment of qualia-based composition, it extends the GL framework in a direction not explicitly formalized there. According to Pustejovsky (Reference Pustejovsky1995: 103), the variables contained in a qualia structure must be fully resolved in the resulting syntactic representation – whether by being syntactically realized, logically bound or existentially closed (a condition he terms ‘Qualia Saturation’). At the same time, his mechanism of Selective Binding allows a modifier to operate on a specific qualia role of the head without altering the overall semantic type of the phrase. The QPP extends these local compositional processes by tracking which qualia roles remain accessible across successive layers of modification after each modifier–head combination. This recursive, resource-sensitive mechanism accounts for the strict ordering constraints among multiple Mandarin de-modifiers, thereby broadening GL’s account of qualia-sensitive modification from individual compositional operations to hierarchically iterated modification.

The QPP exhibits a principled parallel to Baker‑style linking constraints UTAH (Uniformity of Theta Assignment Hypothesis)/RUTAH (Relativized UTAH) (Baker Reference Baker1988, Reference Baker and Haegeman1997), as both govern the mapping of a structured inventory of semantic roles onto syntactic configurations, thereby yielding stable role-configuration mappings. Our proposal recasts this linking logic within the nominal domain as stepwise access to qualia-anchored relation slots under iterative de-linking.

This analogy highlights the resource‑sensitive character of qualia‑based modification: once a qualia relation has been saturated, it is unavailable to subsequent modifiers. Non‑iterable qualia roles therefore block de‑stacking when successive modifiers target the same relation, as in Zhangsan de Lisi de xiaoshuo ‘lit. Zhangsan’s Lisi’s novel’ under the reading where both Zhangsan and Lisi saturate the AGENTIVE role. By contrast, multiple de‑modifications are possible when successive modifiers engage distinct qualia roles remaining accessible within the inherited qualia structure, as in (28). Sections 6.3–6.4 return to a different type of recursive modification, namely those involving gapped relative clauses.

We now turn to the compositional derivation of (28). (30) gives the lexical components of zhaopian ‘photo’ required for semantic composition. To compose (28), selective binding must target the TELIC and AGENTIVE qualia roles to produce the content and agent relations. The result is a three-place predicate with an open event argument. Since no event is required for subsequent nominal composition in this case, EC_event existentially closes the event argument before the derived relation is supplied as a value of R.

(30)	Lexical structure of zhaopian ‘photo’
	(a)	Basic argument structure
		$ [\hskip0.1em [\mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ]=\lambda {x}_e[\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)]. $
	(b)	Selective binding targets TELIC role
		$ {\mathrm{SB}}_{\mathrm{TELIC}}\left(\mathrm{zhaopian}\right)=\lambda {y}_e\lambda {x}_e\lambda {v}_v\left[\begin{array}{c}\mathrm{photo}(x)\wedge \mathrm{depict}(v)\\ {}\wedge \mathrm{Content}\left(v,y\right)\wedge \mathrm{Theme}\left(v,x\right)\end{array}\right]. $
	(c)	ECevent applies to derive the content relation
		$ {R}_{\mathrm{CONTENT}}={\mathrm{EC}}_{\mathrm{event}}({\mathrm{SB}}_{\mathrm{TELIC}}(\mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n})). $
		$ {R}_{\mathrm{CONTENT}}=\lambda {y}_e\lambda {x}_e\mathrm{\exists}{v}_v[\begin{array}{c}\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{e}\mathrm{p}\mathrm{i}\mathrm{c}\mathrm{t}(v)\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{t}(v,y)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(v,x)\end{array}]. $
	(d)	Selective binding targets AGENTIVE role
		$ {\mathrm{SB}}_{\mathrm{AGENTIVE}}(\mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n})=\lambda {a}_e\lambda {x}_e\lambda {v}_v^{\mathrm{\prime}}[\begin{array}{c}\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}({v}^{\mathrm{\prime}})\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}({v}^{\mathrm{\prime}},a)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}({v}^{\mathrm{\prime}},x)\end{array}]. $
	(e)	ECevent applies to derive the agent relation
		$ {R}_{\mathrm{AGENTIVE}}={\mathrm{EC}}_{\mathrm{event}}({\mathrm{SB}}_{\mathrm{AGENTIVE}}(\mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n})). $
		$ {R}_{\mathrm{AGENTIVE}}=\lambda {a}_e\lambda {x}_e\mathrm{\exists}v{{}^{\mathrm{\prime}}}_v[\begin{array}{c}\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}({v}^{\mathrm{\prime}})\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}({v}^{\mathrm{\prime}},a)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}({v}^{\mathrm{\prime}},x)\end{array}]. $

Given these definitions, the semantic composition of (28) proceeds as follows:

(31)	Inner content composition: beijiguang de zhaopian ‘photo of aurora’
	(a)	$ [\hskip0.1em [\mathrm{beijiguang}\hskip0.5em \mathrm{de}]\hskip0.1em ]=[\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ](\mathrm{a}\mathrm{u}\mathrm{r}\mathrm{o}\mathrm{r}\mathrm{a}). $
		$ [\hskip0.1em [\mathrm{beijiguang}\hskip0.5em \mathrm{de}]\hskip0.1em ]=\lambda {R}_{\langle e,\langle e,t\rangle \rangle}\lambda {x}_e[R(\mathrm{a}\mathrm{u}\mathrm{r}\mathrm{o}\mathrm{r}\mathrm{a})(x)]. $
		$ R={R}_{\mathrm{CONTENT}}=\lambda {y}_e\lambda {x}_e\mathrm{\exists}{v}_v[\begin{array}{c}\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{e}\mathrm{p}\mathrm{i}\mathrm{c}\mathrm{t}(v)\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{t}(v,y)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(v,x)\end{array}] $
	(b)	Let $ P=[\hskip0.1em [\mathrm{beijiguang}\hskip0.5em \mathrm{de}\hskip0.5em \mathrm{zhaopian}]\hskip0.1em ]. $
		$ P=[\hskip0.1em [\mathrm{beijiguang}\hskip0.5em \mathrm{de}\hskip0.5em \mathrm{zhaopian}]\hskip0.1em ]=[\hskip0.1em [\mathrm{beijiguang}\hskip0.5em \mathrm{de}]\hskip0.1em ]({R}_{\mathrm{CONTENT}}). $
		$ =\lambda {x}_e\exists {v}_v\left[\mathrm{photo}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{depict}(v)\wedge \mathrm{Content}\left(v,\mathrm{aurora}\right)\hskip0.35em \wedge \hskip0.35em \mathrm{Theme}\left(v,x\right)\right]. $

Per the QPP, the unused AGENTIVE role of the head noun zhaopian remains accessible for further composition. Therefore, selective binding can target this role within the complex nominal constituent beijiguang de zhaopian.

(32)	Outer agent/creator composition
	(a)	SBAGENTIVE applies to beijiguang de zhaopian plus existential closure of ECevent
		$ {\mathrm{EC}}_{\mathrm{event}}({\mathrm{SB}}_{\mathrm{AGENTIVE}}({[\mathrm{b}\mathrm{e}\mathrm{i}\mathrm{j}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n}]}_{zhaopian}))=\lambda {y}_e\lambda {x}_e\mathrm{\exists}{v}_v^{\mathrm{\prime}}\left[\begin{array}{c}P(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}({v}^{\mathrm{\prime}})\\ {}\wedge \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}({v}^{\mathrm{\prime}},y)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}({v}^{\mathrm{\prime}},x)\end{array}\right] $
	(b)	Substituting P yields the following composite relation RAGENTIVE
		$ \lambda {\mathrm{y}}_e\lambda {x}_e[\mathrm{\exists}{v}_v[\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{e}\mathrm{p}\mathrm{i}\mathrm{c}\mathrm{t}(v)\hskip0.35em \wedge \hskip0.35em \mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{t}(v,\mathrm{a}\mathrm{u}\mathrm{r}\mathrm{o}\mathrm{r}\mathrm{a})\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(v,x)] $
		$ \hskip0.35em \wedge \hskip0.35em \mathrm{\exists}{v}_v^{\mathrm{\prime}}[\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}({v}^{\mathrm{\prime}})\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}({v}^{\mathrm{\prime}},y)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}({v}^{\mathrm{\prime}},x)]]. $
	(c)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]=\lambda {R}_{\langle e,\langle e,t\rangle \rangle}\lambda {x}_e[R(\mathrm{Z}\mathrm{S})(x)]. $
	(d)	Final composition: substitution of R in (32c) with RAGENTIVE
		$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{b}\mathrm{e}\mathrm{i}\mathrm{j}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{a}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ]=[\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]({R}_{\mathrm{AGENTIVE}}) $
		$ =\lambda {x}_e\Big[\exists {v}_v\left[\mathrm{photo}(x)\wedge \mathrm{depict}(v)\wedge \mathrm{Content}\left(v,\mathrm{aurora}\right)\wedge \mathrm{Theme}\left(v,x\right)\right] $
		$ \hskip0.35em \wedge \hskip0.35em \mathrm{\exists}{v}_v^{\mathrm{\prime}}[\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}({v}^{\mathrm{\prime}})\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}({v}^{\mathrm{\prime}},\mathrm{Z}\mathrm{S})\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}({v}^{\mathrm{\prime}},x)]]. $

Thus, the final expression denotes a photo depicting the aurora taken by Zhangsan. This derivation preserves both distinct event dependencies while ensuring each qualia relation is accessed only once and remains anchored in the lexical head zhaopian.

5.4. Predicate-oriented AP modifiers and functional-application resolution

This section examines AP modifiers frequently characterized as non-restrictive, non-predicative or intensional such as weiyi ‘only/unique’, zhuyao ‘main’ and qianren ‘former’. Unlike ordinary intersective adjectives, these elements do not denote first-order individual properties. Instead, they operate as predicate modifiers, taking a nominal predicate as input and returning a modified predicate. They can determine the cardinality of the predicate’s contextual extension (weiyi), impose an ordering over alternatives (zhuyao) or shift its temporal profile (qianren). Recognizing this type distinction allows these modifiers to integrate seamlessly into the single-linker architecture without departing from strict bottom-up composition. As the following analysis demonstrates, while the AP introduces these specific interpretive effects, de maintains a uniform semantics as a polymorphic linker. Consequently, the non-restrictive or intensional profile is entirely internal to the AP, eliminating the need for a distinct semantic entry for de.

Recall the core linker schema in (13). The semantic type of the pre-de modifier strictly constrains which admissible relation can instantiate R. In ordinary intersective modification, a first-order modifier of type ⟨e,t⟩ combines with the current first-order, property-denoting nominal constituent, also of type ⟨e,t⟩, via the Predicate-Modification Interface relation (defined in Section 6.3.2) to yield a property of individuals. However, predicate-oriented APs such as weiyi ‘only/unique’, zhuyao ‘main’ and qianren ‘former’ are higher-order expressions of type ⟨⟨e,t⟩, ⟨e,t⟩⟩: they take a property as input and return a modified property. Standard predicate modification cannot apply in this context, as it requires two arguments of matching first-order types rather than a functor and its argument. A qualia-instantiated relation is similarly unsuitable, as it interprets the modifier as saturating a noun-internal qualia role; the APs in question instead require the nominal predicate itself as their semantic argument. Therefore, we propose that de provides a grammar-driven interface in these contexts, linking a predicate-oriented AP to a nominal constituent via the functional-application relation defined in (34).

(34)	Functional-Application Interface Relation
	(a)	$ {R}_{\mathrm{Fapp},N}=\unicode{x03BB} {M}_{\langle \langle e,t\rangle, \langle e,t\rangle \rangle}\unicode{x03BB} {x}_e.M([\hskip0.1em [N]\hskip0.1em ])(x) $ Type: $ \left\langle \left\langle \left\langle e,t\right\rangle, \left\langle e,t\right\rangle \right\rangle, \left\langle e,t\right\rangle \right\rangle $
	(b)	$ {R}_{\mathrm{Fapp},N}\in \mathrm{SourceR}\left({N}_{\mathrm{Intf}}\right) $

In this configuration, there is no type mismatch between the predicate-oriented modifier M and the nominal predicate N; M can directly take N as its argument. When de is present, it first takes M as its argument, and the interface relation R_Fapp,N simply facilitates this standard functional application, applying M as a functor over N to yield an individual property. In this capacity, de functions as a pure syntactic and semantic linker. Because M and N can compose directly without mediation, this correctly predicts that de is optional for these modifiers.

(35)	$ [\hskip0.1em [M\hskip0.32em de]\hskip0.1em ]=[\hskip0.1em [de]\hskip0.1em ]([\hskip0.1em [M]\hskip0.1em ])=\lambda {R}_{\langle \langle \langle e,t\rangle, \langle e,t\rangle \rangle, \langle e,t\rangle \rangle}\lambda {x}_e.\hskip0.1em R\hskip0.1em ([\hskip0.1em [M]\hskip0.1em ])(x) $

To illustrate, let liyou ‘reason’ denote a simple property: $ [\hskip0.1em [\mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}]\hskip0.1em ]=\lambda {x}_e.\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{o}\mathrm{n}(x) $ . For present purposes, we represent weiyi ‘only/unique’ as a predicate modifier that carries a contextually relevant uniqueness presupposition:

(36)	$ {[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}]\hskip0.1em ]}^c=\lambda {P}_{\langle e,t\rangle }:\mathrm{\exists}!x[P(x)\hskip0.35em \wedge \hskip0.35em C(x)].\lambda x[P(x)\hskip0.35em \wedge \hskip0.35em C(x)] $ ,

where C is the contextually supplied restriction.

When weiyi de merges with the head noun liyou, the underspecified relation R resolves to the type-compatible interface relation, R_Fapp,N. Semantically, this allows the AP to apply directly to the head predicate ⟦liyou⟧, yielding an ordinary predication on individuals enriched with a uniqueness presupposition.

(37)	(a)	$ {[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}]\hskip0.1em ]}^c=[[\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ]({[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}]\hskip0.1em ]}^c)]({R}_{\mathrm{Fapp},\mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}}) $
		$ =\lambda x.{R}_{\mathrm{Fapp},\mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}}({[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}]\hskip0.1em ]}^c)(x) $
		$ =\lambda x.{[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}]\hskip0.1em ]}^c([\hskip0.1em [\mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}]\hskip0.1em ])(x) $
	(b)	Since $ [\hskip0.1em [\mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}]\hskip0.1em ]=\lambda {x}_e.\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{o}\mathrm{n}(x) $ , (36) yields:
		$ {[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}]\hskip0.1em ]}^c([\hskip0.1em [\mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}]\hskip0.1em ])=\lambda x[\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{o}\mathrm{n}(x)\hskip0.35em \wedge \hskip0.35em C(x)] $ ,
		defined only if $ \exists !x\left[\mathrm{reason}(x)\wedge C(x)\right] $
	(c)	$ {[\hskip0.1em [\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{y}\mathrm{i}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{l}\mathrm{i}\mathrm{y}\mathrm{o}\mathrm{u}]\hskip0.1em ]}^c $ = $ \lambda x\left[\mathrm{reason}(x)\wedge C(x)\right] $ ,
		defined only if $ \exists !x\left[\mathrm{reason}(x)\wedge C(x)\right] $

This derivation demonstrates that the uniqueness interpretation is achieved contextually, avoiding the need to posit a separate non-restrictive or intensional meaning for de. Importantly, just as in previous derivations, the restriction on the individual variable originates entirely from the selected relation R (here, the N-indexed R_Fapp,N).Footnote ⁷ This same predicate-oriented, type-driven resolution naturally extends to other modifiers like zhuyao ‘main’ (ordering-based) and qianren ‘former’ (temporal), with their respective semantic parameters made explicit as required.

6.1. Propositional content clauses

Consider again (3), Zhangsan pochan de yaoyan ‘the rumor that Zhangsan went bankrupt’, where the clausal modifier provides the head noun’s propositional content. Following Kratzer (Reference Kratzer2006), Moulton (Reference Moulton2009, Reference Moulton2015), Moltmann (Reference Moltmann2013) and Lin (Reference Lin2024), we analyze content nouns like rumor as denoting content individuals, or proposition-bearing objects. In our framework, selective binding establishes a constitutive content relation between a proposition and the proposition-bearing object, allowing the clausal modifier to saturate the proposition argument of that relation.

The relevant lexical information is as follows:

(38)	(a)	$ [\hskip0.1em [\mathrm{y}\mathrm{a}\mathrm{o}\mathrm{y}\mathrm{a}\mathrm{n}]\hskip0.1em ]=\lambda {x}_{\mathrm{e}}.\mathrm{r}\mathrm{u}\mathrm{m}\mathrm{o}\mathrm{r}(x) $
	(b)	$ {\mathrm{SB}}_{\mathrm{CONSTITUTIVE}}\left(\mathrm{yaoyan}\right) $
		$ =\lambda {p}_{\left\langle s,t\right\rangle}\;\lambda {x}_e.\left[\mathrm{rumor}(x)\wedge {\mathrm{constitute}}_{\mathrm{yaoyan}}\left(p,x\right)\right] $

In (38b), selectively binding the head noun’s CONSTITUTIVE role yields a role-instantiated relation, mapping a given proposition p to the property of being a rumor whose content is p. Here, de links the propositional modifier to the head-licensed relation. The general schema for de in propositional content clauses, along with an illustrative derivation, is provided below:

(39)	$ [\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ] $ (for propositional clauses)
	$ [\hskip0.1em [{\mathrm{de}}_{\mathrm{pro}.\mathrm{c}\mathrm{l}}]\hskip0.1em ]=\lambda {p}_{\langle s,t\rangle}\hskip0.5em \lambda {R}_{\langle \langle s,t\rangle, \langle e,t\rangle \rangle}\hskip0.5em \lambda {x}_e.R(p)(x) $

(40)	Semantic composition of (3)
	(a)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{p}\mathrm{o}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]=[\hskip0.1em [{\mathrm{de}}_{\mathrm{pro}.\mathrm{c}\mathrm{l}}]\hskip0.1em ]([\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{p}\mathrm{o}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}]\hskip0.1em ]) $
		$ =\lambda R\lambda x.R\left(\lambda w\left[\mathrm{bankrupt}\left(\mathrm{ZS},w\right)\right]\right)(x) $
	(b)	$ R={R}_{\mathrm{Yaoyan}}={\mathrm{SB}}_{\mathrm{CONSTITUTIVE}}\left(\mathrm{yaoyan}\right) $
		R $ =\lambda {p}_{\left\langle s,t\right\rangle}\;\lambda {x}_e.\left[\mathrm{rumor}(x)\wedge {\mathrm{constitute}}_{\mathrm{yaoyan}}\left(p,x\right)\right] $
	(c)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{p}\mathrm{o}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{y}\mathrm{a}\mathrm{o}\mathrm{y}\mathrm{a}\mathrm{n}]\hskip0.1em ] $
		$ =\lambda {x}_e.\left[\mathrm{rumor}(x)\wedge {\mathrm{constitute}}_{\mathrm{yaoyan}}\left(\lambda w\left[\mathrm{bankrupt}\left(\mathrm{ZS},w\right)\right],x\right)\right] $

This final interpretation yields the property of being a rumor whose constitutive content is the proposition that Zhangsan went bankrupt.

Consequently, unlike complementizer-based or type-shifting accounts that underspecify the clause-noun relation, our relation-variable architecture maintains a uniform linker and derives the content interpretation directly from the head noun.

6.2. Saturated event clauses

6.2.2. Baseline: event-individual modification with an inherently relational noun

To establish a baseline template, we begin with an inherently relational event noun. Following Huang (Reference Huang2016), Kou & Yuan (Reference Kou and Yuan2017) and Lin (Reference Lin2024), the noun jieguo ‘result’ lexically specifies a causal relation mapping events to entities:

(43)

$ [\hskip0.1em [\mathrm{j}\mathrm{i}\mathrm{e}\mathrm{g}\mathrm{u}\mathrm{o}]\hskip0.1em ]=\lambda {v}_{\mathrm{v}}\hskip0.5em \lambda {x}_{\mathrm{e}}.[\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{u}\mathrm{l}\mathrm{t}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{c}\mathrm{a}\mathrm{u}\mathrm{s}\mathrm{e}(v,x)] $

In the noun phrase zhe ci shijian de jieguo ‘the result of this event’, the modifier zhe ci shijian ‘this event’ denotes a contextually identified event individual. Let this individual be represented by v_c. The linker de connects this specific event to the inherently relational head noun jieguo.

(44)	(a)	$ [\hskip0.1em [{\mathrm{de}}_{\mathrm{e}\mathrm{vent}}]\hskip0.1em ]=\lambda {v}_v\hskip0.5em \lambda {R}_{\langle v,\langle e,t\rangle \rangle}\hskip0.5em \lambda {x}_{\mathrm{e}}.[R(v)(x)] $
	(b)	Applying de to the specific event:
		$ [\hskip0.1em [\mathrm{z}\mathrm{h}\mathrm{e}\hskip0.5em \mathrm{c}\mathrm{i}\hskip0.5em \mathrm{s}\mathrm{h}\mathrm{i}\mathrm{j}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]=\lambda {R}_{\langle v,\langle e,t\rangle \rangle}\hskip0.5em \lambda {x}_e.[R({v}_c)(x)] $
	(c)	Combining with the head noun jieguo:
		$ [\hskip0.1em [\mathrm{z}\mathrm{h}\mathrm{e}\hskip0.5em \mathrm{c}\mathrm{i}\hskip0.5em \mathrm{s}\mathrm{h}\mathrm{i}\mathrm{j}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{j}\mathrm{i}\mathrm{e}\mathrm{g}\mathrm{u}\mathrm{o}]\hskip0.1em ]=\lambda x.[\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{u}\mathrm{l}\mathrm{t}(x)\hskip0.35em \wedge \hskip0.35em \mathrm{c}\mathrm{a}\mathrm{u}\mathrm{s}\mathrm{e}({v}_c,x)] $

This derivation requires neither coercion nor selective binding. Because the head noun is inherently relational, its lexically encoded event-result relation can serve directly as an admissible value for the relational variable R.

6.2.3. SEC composition and event-property-to-event-individual coercion

We now turn to the noun shengyin ‘sound’, which we analyze as an auditory effect token associated with an event. Unlike jieguo ‘result’, shengyin need not be relational in its basic predication (cf. you shengyin ‘there is a sound’). Nevertheless, our framework allows this non-relational basic predication to remain compatible with a head-licensed linking option. Specifically, the noun can still supply an AGENTIVE-derived source/effect relation within its qualia structure to allow for de-linking through selective binding. Accordingly, we assume the basic argument structure for shengyin is represented in (45). Selective binding into the AGENTIVE role then yields the relation shown in (46).

(45)	$ [\hskip0.1em [s\mathrm{h}\mathrm{e}\mathrm{n}\mathrm{g}\mathrm{y}\mathrm{i}\mathrm{n}]\hskip0.1em ]=\lambda x.\mathrm{s}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{d}(x) $

(46)	$ {R}_{\mathrm{Shengyin}}={\mathrm{SB}}_{\mathrm{AGENTIVE}}\left(\mathrm{shengyin}\right)=\lambda {v}_{\mathrm{v}}\;\lambda {x}_{\mathrm{e}}.\left[\mathrm{sound}(x)\wedge \mathrm{cause}\left(v,x\right)\right] $

This relation provides an admissible value for de’s underspecified relation variable R. The remaining issue arises from the modifier side.

Following Section 6.2.1, SECs denote event predicates (type ⟨v,t⟩; Lin Reference Lin2024), rather than event individuals, meaning the SEC Zhangsan tan gangqin ‘Zhangsan plays the piano’ denotes the event property in (47).

(47)

$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{tan}\hskip0.5em \mathrm{g}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{q}\mathrm{i}\mathrm{n}]\hskip0.1em ]=\lambda {v}_{\mathrm{v}}.[\mathrm{\exists}y[\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{o}(y)\hskip0.35em \wedge \hskip0.35em \mathrm{p}\mathrm{l}\mathrm{a}\mathrm{y}(v,\mathrm{Z}\mathrm{S},y)]] $

However, this event property cannot serve as the argument of the AGENTIVE-licensed source/effect relation in (46), because the relation requires an event individual for its modifier-side argument, whereas (47) provides an event property. As a result, establishing a type-compatible value for R requires shifting the SEC from an event property to an event individual. This genuine, independently motivated type mismatch triggers coercive type shifting in the narrow sense defined in Section 4. To repair the mismatch, a contextually restricted property-to-individual shift isolates a contextually identified event token. We model this coercion for episodic construals using the familiar iota-type shift (cf. Partee Reference Partee, Groenendijk, de Jongh and Stokhof1987).

(48)	(a)	$ {\sigma}_{C_v}(P)=\iota v\left[P(v)\wedge {C}_v(v)\right] $ $ \left(\mathrm{with}\hskip0.32em P\;\mathrm{of}\ \mathrm{type}\;\left\langle v,t\right\rangle \right) $
	(b)	$ {\sigma}_{C_v}\left(\lambda v\left[\exists y\left[\mathrm{piano}(y)\wedge \mathrm{play}\left(v,\mathrm{ZS},y\right)\right]\right]\right) $
		$ =\iota v\left[\exists y\left[\mathrm{piano}(y)\wedge \mathrm{play}\left(v,\mathrm{ZS},y\right)\right]\wedge {C}_v(v)\right] $

In (48), C_v is a contextually supplied restriction on events; the ɩ-term is defined only if a unique event satisfies both P and C_v. With the contextually restricted ɩ-operation applied, de composes exactly as in Section 6.2.2:

(49)	(a)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{tan}\hskip0.5em \mathrm{g}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{q}\mathrm{i}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ] $
		$ =[\hskip0.1em [{\mathrm{de}}_{\mathrm{event}}]\hskip0.1em ]({\sigma}_{C_v}([\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{tan}\hskip0.5em \mathrm{g}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{q}\mathrm{i}\mathrm{n}]\hskip0.1em ])) $
		$ =\lambda R\;\lambda x.\left[R\left(\iota v\left[\exists y\left[\mathrm{piano}(y)\wedge \mathrm{play}\left(v,\mathrm{ZS},y\right)\right]\wedge {C}_v(v)\right]\right)(x)\right] $
	(b)	$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{tan}\hskip0.5em \mathrm{g}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{q}\mathrm{i}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{s}\mathrm{h}\mathrm{e}\mathrm{n}\mathrm{g}\mathrm{y}\mathrm{i}\mathrm{n}]\hskip0.1em ] $
		$ =\lambda x.\left[\mathrm{sound}(x)\wedge \mathrm{cause}\left(\iota v\left[\exists y\left[\mathrm{piano}(y)\wedge \mathrm{play}\left(v,\mathrm{ZS},y\right)\right]\wedge {C}_v(v)\right],x\right)\right] $

This denotes the set of sound tokens produced by Zhangsan’s contextually identified piano-playing event.

SECs may also receive habitual or generic interpretations in appropriate discourse contexts. Such readings involve independent generic quantification over events or situations and do not affect our central point here; we therefore set aside the detailed interaction between generic quantification and nominal composition.

6.3. Gapped relative clauses

6.3.2. The Predicate-Modification Interface

Restrictive relative clauses and ordinary intersective adjectives are standardly analyzed as individual properties of type ⟨e,t⟩. They combine via predicate modification, yielding their intersection (Heim & Kratzer Reference Heim and Kratzer1998). We propose that this independently motivated compositional operation serves as a grammar-given interface value for R. Let Nom denote the current property-denoting nominal constituent of type ⟨e,t⟩, headed by the lexical N. This interface relation is indexed to the current Nom constituent, as given below:

(50)	predicate-modification interface Rpm,Nom
	$ {R}_{\mathrm{pm},\mathrm{N}\mathrm{o}\mathrm{m}}=\lambda {P}_{\langle e,t\rangle}\hskip0.5em \lambda x.[[\hskip0.1em [Nom]\hskip0.1em ](x)\hskip0.35em \wedge \hskip0.35em P(x)] $ Type: $ \left\langle \left\langle e,t\right\rangle, \left\langle e,t\right\rangle \right\rangle $
	$ {R}_{\mathrm{pm},\mathrm{Nom}}\in \mathrm{SourceR}\left({N}_{\mathrm{Intf}}\right) $

In (50), we use Nom rather than N because Predicate Modification may apply to a nominal constituent larger than N. Unlike a qualia-instantiated relation, R_pm,Nom directly implements intersective Predicate Modification rather than expressing a substantive lexical link between a modifier and a nominal. Because R_pm,Nom already contains the nominal constituent’s denotation, resolving R to this relation directly yields the fully modified property.

To illustrate, consider the noun phrase piaoliang de qunzi ‘beautiful skirt’:

(51)	(a)	$ [\hskip0.1em [\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{g}]\hskip0.1em ]=\lambda x.\mathrm{b}\mathrm{e}\mathrm{a}\mathrm{u}\mathrm{t}\mathrm{i}\mathrm{f}\mathrm{u}\mathrm{l}(x) $
	(b)	Apply the general linker:
		$ [\hskip0.1em [\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]=\lambda {R}_{\langle \langle e,t\rangle, \langle e,t\rangle \rangle}\hskip0.5em \lambda {x}_e.R([\hskip0.1em [piaoliang]\hskip0.1em ])(x) $
	(c)	Resolve R to R pm,qunzi
		$ [\hskip0.1em [\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]({R}_{\mathrm{pm},\mathrm{q}\mathrm{u}\mathrm{n}\mathrm{z}\mathrm{i}})=\lambda x.{R}_{\mathrm{pm},\mathrm{q}\mathrm{u}\mathrm{n}\mathrm{z}\mathrm{i}}([\hskip0.1em [\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{g}]\hskip0.1em ])(x) $
		$ =\lambda x[[\hskip0.1em [\mathrm{q}\mathrm{u}\mathrm{n}\mathrm{z}\mathrm{i}]\hskip0.1em ](x)\hskip0.35em \wedge \hskip0.35em [\hskip0.1em [\mathrm{p}\mathrm{i}\mathrm{a}\mathrm{o}\mathrm{l}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{g}]\hskip0.1em ](x)] $
		$ =\lambda x\left[\mathrm{skirt}(x)\wedge \mathrm{beautiful}(x)\right] $

This yields an ordinary intersective interpretation. Importantly, this analysis does not imply that all Predicate Modification in Mandarin requires the presence of de, as de-less modification is well-attested in the right contexts. Rather, it demonstrates that within the specific de-marked construction under investigation, predicate modification is achieved by resolving the de-introduced relation variable to R_pm,Nom. The primary function of R_pm,Nom is therefore analogous to R_Fapp,N, which is required for the predicate-oriented modification discussed earlier.

6.3.3. Derivation of modification by GRC

The derivation of a nominal constituent modified by a gapped relative clause (GRC), such as Zhangsan zhaodao de gongju ‘the tool that Zhangsan found’, parallels the derivation of the noun phrase piaoliang de qunzi ‘beautiful skirt’ from the previous subsection. We assume that a restrictive clause is derived by movement (e.g. Huang Reference Huang1982) and that abstraction over the object gap independently yields an individual property of type ⟨e,t⟩ (Heim & Kratzer Reference Heim and Kratzer1998). A restrictive relative thus provides precisely the appropriate input for the Predicate-Modification Interface defined in (50). Example (52) illustrates this semantic composition:

(52)	(a)	Relative clause property
		$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{d}\mathrm{a}\mathrm{o}\_]\hskip0.1em ]=\lambda x.\mathrm{\exists}e[\mathrm{f}\mathrm{i}\mathrm{n}\mathrm{d}(e)\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(e,\mathrm{Z}\mathrm{S})\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)] $
	(b)	Apply the general linker
		$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{d}\mathrm{a}\mathrm{o}\_\mathrm{d}\mathrm{e}]\hskip0.1em ]=[\hskip0.1em [de]\hskip0.1em ]([\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{d}\mathrm{a}\mathrm{o}\_]\hskip0.1em ]) $
		$ =\lambda {R}_{\langle \langle e,t\rangle, \langle e,t\rangle \rangle}\hskip0.5em \lambda {x}_e.R([\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{d}\mathrm{a}\mathrm{o}\_]\hskip0.1em ])(x) $
	(c)	Resolve R to $ {R}_{\mathrm{pm},\mathrm{gongju}} $
		$ [\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{d}\mathrm{a}\mathrm{o}\_\mathrm{d}\mathrm{e}]\hskip0.1em ]({R}_{\mathrm{pm},\mathrm{g}\mathrm{o}\mathrm{n}\mathrm{g}\mathrm{j}\mathrm{u}}) $
		$ =\lambda x.{R}_{\mathrm{pm},\mathrm{g}\mathrm{o}\mathrm{n}\mathrm{g}\mathrm{j}\mathrm{u}}([\hskip0.1em [\mathrm{Z}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{d}\mathrm{a}\mathrm{o}\_]\hskip0.1em ])(x) $
		$ =\lambda x\left[\mathrm{tool}(x)\wedge \exists e\left[\mathrm{find}(e)\wedge \mathrm{Agent}\left(e,\mathrm{ZS}\right)\wedge \mathrm{Theme}\left(e,x\right)\right]\right] $

Significantly, the current architecture achieves this without analyzing de as semantically vacuous. Instead, de introduces the constant relation variable R, which is subsequently resolved to the independently motivated Predicate-Modification Interface, R_pm,Nom.

In the simple case presented in (52), Nom corresponds directly to the property denoted by the noun gongju ‘tool’. However, the broader Nom-based formulation proves essential in stacked configurations such as those discussed in the next section, where both a GRC and inner NCC or SEC occur.

6.4. Ordering constraints among prenominal clauses: a compositional account

Huang (Reference Huang2016) points out that, when a head noun is modified by multiple prenominal clauses, gapped relative clauses (GRCs) systematically precede noun complement clauses (NCCs) and saturated event clauses (SECs). The contrast in (53)–(54) illustrates this with (b) orders degraded relative to (a) orders.

These ordering facts derive naturally from the independently motivated distinction between the nominal head’s lexical (or selectional) domain and adjunct-level modification. Because NCCs and SECs saturate relations licensed by the lexical head, they compose directly within its selectional domain. Conversely, GRCs do not saturate a lexical relation; following abstraction over the gap, they combine through the Predicate-Modification Interface with the nominal property constructed by this inner domain. (55) provides a compact schematic derivation of (53a):

(55)	(a)	The NCC first combines with the head
		$ {Nom}_1=\lambda x\;\left[\mathrm{question}(x)\wedge {\mathrm{constitute}}_{\mathrm{wenti}}\left(p,x\right)\right] $ ,
		where $ p=[\hskip0.1em [\mathrm{y}\mathrm{a}\mathrm{o}-\mathrm{b}\mathrm{u}-\mathrm{y}\mathrm{a}\mathrm{o}\hskip0.5em \mathrm{k}\mathrm{u}\mathrm{o}\mathrm{d}\mathrm{a}\ \mathrm{z}\mathrm{h}\mathrm{a}\mathrm{o}\mathrm{s}\mathrm{h}\mathrm{e}\mathrm{n}\mathrm{g}]\hskip0.1em ] $ (whether to expand enrollment)
	(b)	The denotation of outer GRC
		$ Q=[\hskip0.1em [\mathrm{w}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{e}\mathrm{n}\mathrm{g}\mathrm{z}\mathrm{a}\mathrm{i}\ \mathrm{k}\mathrm{a}\mathrm{o}\mathrm{l}\ddot{\mathrm{u}}\_]\hskip0.1em ] $
		$ =\lambda x.\exists e\left[\mathrm{consider}(e)\wedge \mathrm{Agent}\left(e,\mathrm{we}\right)\wedge \mathrm{Theme}\left(e,x\right)\right] $
	(c)	Outer GRC combines with Nom1 via the Predicate Modification Interface.
		$ [\hskip0.1em [\mathrm{w}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{e}\mathrm{n}\mathrm{g}\mathrm{z}\mathrm{a}\mathrm{i}\ \mathrm{k}\mathrm{a}\mathrm{o}\mathrm{l}\ddot{\mathrm{u}}\_\mathrm{d}\mathrm{e}]\hskip0.1em ]({R}_{\mathrm{pm},{\mathrm{Nom}}_1}) $
		$ =\lambda x.{R}_{\mathrm{pm},{\mathrm{Nom}}_1}([\hskip0.1em [\mathrm{w}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{n}\ \mathrm{z}\mathrm{h}\mathrm{e}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{z}\mathrm{a}\mathrm{i}\hskip0.5em \mathrm{k}\mathrm{a}\mathrm{o}\mathrm{l}\ddot{\mathrm{u}}\_]\hskip0.1em ])(x) $
		$ =\lambda x.{R}_{\mathrm{pm},{\mathrm{Nom}}_1}(Q)(x) $
		$ =\lambda x\hskip0.5em [\begin{array}{c}\mathrm{q}\mathrm{u}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}(x)\hskip0.35em \wedge \hskip0.35em {\mathrm{constitute}}_{\mathrm{wenti}}(p,x)\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{\exists}e\hskip0.5em [\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{d}\mathrm{e}\mathrm{r}(e)\hskip0.35em \wedge \hskip0.35em \mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(e,\mathrm{w}\mathrm{e})\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)]\end{array}] $

The outcome of (55) represents the property of being a question x whose content is p (namely, whether to expand enrollment) and which we are considering.

The case of (54a) is analogous, with Nom₁ instantiated by the property produced by the inner SEC-head combination. We will not repeat the details.

The restriction in (53) and (54) reflects the well-known complement-adjunct, or local-licensing, asymmetry in Chinese nominals (Huang Reference Huang2016): head-dependent complements merge structurally closer to the nominal head than restrictive relative clauses do. Importantly, this restriction does not follow from semantic type alone. If Predicate Modification Interface could apply first while lexical relations remained freely saturable at higher structural levels, nothing would rule out a derivation in which a GRC first yields a modified property, and an NCC or SEC subsequently saturates a head-licensed relation.

To resolve this, we propose that such derivations are excluded not by type mismatches, but because the relations employed by de obey a strict hierarchical priority: lexical saturation must precede non-lexical modification. Mechanistically, while qualia-based selective binding preserves unbound qualia roles and passes them upward via the QPP, grammar-driven interface relations (such as R_pm,Nom) structurally block the upward transmission of this qualia structure.Footnote ⁸ Consequently, the application of an adjunct interface relation irreversibly closes the noun’s lexical/selectional domain, effectively preventing higher nodes from accessing the QPP framework to satisfy any remaining lexical requirements.

7.1. Duration pseudo-modification via a telic event profile and VP-level Event-dependent Restrict

As noted, in duration pseudo-modification, a DP-internal measure phrase appears to modify the matrix event, giving rise to an apparent syntax-semantics mismatch. For example, in (56), the nominal constituent san tian de ge forms an object DP before combining with the verb.

A duration phrase conventionally denotes an event predicate modifier of type ⟨⟨v,t⟩,⟨v,t⟩⟩, as independently motivated by its use in verbal modification in (57) (Lin Reference Lin, Dölling, Heyde-Zybatow and Schäfer2008).

In (57), the duration phrase takes the event property denoted by the verb as its argument and returns a property restricted to events whose runtime satisfies the relevant duration condition. Thus, the denotation of san tian ‘three days’ is (58).

(58)

$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\ \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ]=\lambda {P}_{\langle v,t\rangle}\lambda {e}_v\hskip0.5em [P(e)\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}(e)=3\cdot \mathrm{d}\mathrm{a}\mathrm{y}\mathrm{s}] $

Assuming this same denotation applies to the nominal modification in (56), a type mismatch arises because a duration cannot take an individual argument. However, selective binding can access the lexical head’s TELIC role, making an event argument available for duration modification. Since the duration constituent lacks a value for the role-filler argument u, this argument is closed by EC. The nominal argument x and the event argument e remain open.

(59)	Selective binding of the TELIC role of ge ‘song’ plus ECouter
	$ {T}_{\mathrm{telic},\mathrm{ge}}:= {\mathrm{EC}}_{\mathrm{outer}}\left({\mathrm{SB}}_{\mathrm{TELIC}}\left(\mathrm{ge}\right)\right) $
	$ {\mathrm{EC}}_{\mathrm{outer}}\left({\mathrm{SB}}_{\mathrm{TELIC}}\left(\mathrm{ge}\right)\right)=\unicode{x03BB} {x}_e\unicode{x03BB} {e}_v\exists {u}_e\left[{\mathrm{SB}}_{\mathrm{TELIC}}\left(\mathrm{ge}\right)(u)(x)(e)\right] $
	$ =\unicode{x03BB} {x}_e\unicode{x03BB} {e}_v\left[{\mathrm{song}}^{\star }(x)\wedge {\mathrm{sing}}^{\star }(e)\wedge \mathrm{Theme}\left(e,x\right)\wedge \exists {u}_e\left[\mathrm{Agent}\left(e,u\right)\right]\right]. $

In (59), the telic singing relation characterizes a (potentially plural) song as the theme of a pluralizable singing event. Since EC_outer targets only the implicit performer argument, the nominal and event arguments remain available for further composition.

The linker must now connect the higher-order duration modifier in (58) to the nominal argument by assigning a value to R. Because the TELIC-derived relation cannot accept the duration modifier directly as a role filler, we introduce a type-driven interface relation to apply the modifier directly to the head noun’s underlying event predicate. As in Section 5.4, de initially composes with the modifier, leaving R unvalued until resolved by merging the lexical nominal head.

(60)	Duration-TELIC application interface schema
	$ {R}_{\mathrm{Dur}-\mathrm{telic},N}=\lambda {D}_{\left\langle \left\langle v,t\right\rangle, \left\langle v,t\right\rangle \right\rangle}\lambda {x}_e\lambda {e}_v\left[D\left({T}_{\mathrm{telic},N}(x)\right)(e)\right] $ Type: ⟨⟨⟨v,t⟩,⟨v,t⟩⟩,⟨e,⟨v,t⟩⟩⟩

Under the general linker semantics in (13), the Duration-Telic interface relation in (60) serves as a valid instantiation of R. Here, τ = ⟨⟨v,t⟩,⟨v,t⟩⟩ and ρ = ⟨v,t⟩. After the duration modifier D and the interface relation R_Dur-telic,N are integrated, the resulting expression is a function from nominal individuals to event predicates, typed as ⟨e,⟨v,t⟩⟩, as shown in (61).

(61)	Duration-modifier instantiation of (13)
	$ [\hskip0.1em [\mathrm{d}{\mathrm{e}}_{\mathrm{dur}}]\hskip0.1em ]=\lambda {D}_{\langle \langle v,t\rangle, \langle v,t\rangle \rangle}\hskip0.5em \lambda {R}_{\langle \langle \langle v,t\rangle, \langle v,t\rangle \rangle, \langle e,\langle v,t\rangle \rangle \rangle}\hskip0.5em \lambda {x}_e.R(D)(x) $ , where R(D)(x) is type ⟨v,t⟩.

Applying (61) to san tian de ge in (56) yields the derivation in (62).

(62)	(a)	$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{g}\mathrm{e}]\hskip0.1em ]=[\hskip0.1em [\mathrm{d}{\mathrm{e}}_{\mathrm{dur}}]\hskip0.1em ]([\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ])({R}_{\mathrm{dur}-\mathrm{t}\mathrm{e}\mathrm{l}\mathrm{i}\mathrm{c},\mathrm{g}\mathrm{e}}). $
	(b)	By the definition of the Duration-TELIC interface schema in (60)
		$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{g}\mathrm{e}]\hskip0.1em ]=\lambda {x}_e.{R}_{\mathrm{dur}-\mathrm{t}\mathrm{e}\mathrm{l}\mathrm{i}\mathrm{c},\mathrm{g}\mathrm{e}}([\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ])(x) $
		$ =\lambda {x}_e[\lambda {D}_{\langle \langle v,t\rangle, \langle v,t\rangle \rangle}\lambda {y}_e\lambda {e}_v.\operatorname{}D({\mathrm{T}}_{\mathrm{telic},\mathrm{g}\mathrm{e}}(y))(e)]([\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ])(x)\operatorname{} $
		$ =\lambda {x}_e.\lambda {e}_v[[\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}]\hskip0.1em ]({T}_{\mathrm{telic},\mathrm{g}\mathrm{e}}(x))(e)] $
	(c)	Substituting the duration denotation in (58) and the relation Ttelic,ge from (59)
		$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{g}\mathrm{e}]\hskip0.1em ]=\lambda {x}_e\lambda {e}_v[\begin{array}{c}{\mathrm{song}}^{\star }(x)\hskip0.35em \wedge \hskip0.35em {\mathrm{sing}}^{\star }(e)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{\exists}{u}_e[\mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(e,u)]\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}(e)=3\cdot \mathrm{d}\mathrm{a}\mathrm{y}\mathrm{s}\end{array}]. $

The resulting duration-modified nominal constituent now denotes an event-dependent relation over a nominal participant and an event, of type ⟨e,⟨v,t⟩⟩.

Assuming Kratzer’s (Reference Kratzer, Rooryck and Zaring1996) theory of Voice introducing the external argument, we assign the transitive verb chang ‘sing’ a neo-Davidsonian denotation of type ⟨e,⟨v,t⟩⟩, matching the duration-modified nominal constituent in (62c). Since both denote relations of type ⟨e,⟨v,t⟩⟩, neither can take the other as an argument, blocking standard Functional Application. To resolve this and facilitate composition, we introduce the Event-dependent Restrict rule (abbreviated as EDR) in (63). This mechanism synthesizes three established operations: Higginbotham’s (Reference Higginbotham1985) Theta-Identification (equating two open theta roles), Kratzer’s (Reference Kratzer, Rooryck and Zaring1996) Event Identification (identifying the event arguments of Voice and VP) and Chung & Ladusaw’s (Reference Chung and Ladusaw2004) Restrict (restricting a verb’s internal argument with a nominal property without closing the variable). EDR identifies the corresponding nominal and event arguments of the two event-dependent relations while leaving the nominal argument unsaturated.

(63)	event-dependent restrict (EDR)
	$ \mathrm{EDR}\left({f}_{\left\langle e,\left\langle v,t\right\rangle \right\rangle },{g}_{\left\langle e,\left\langle v,t\right\rangle \right\rangle}\right)\to \lambda {x}_e\lambda {e}_v\;\left[f(x)(e)\hskip0.35em \wedge \hskip0.35em g(x)(e)\right] $

With (63), the derivation of (56) proceeds as follows.

(64)	(a)	Duration-modified object DP denotation
		$ [\hskip0.1em [\mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{g}\mathrm{e}]\hskip0.1em ]=\unicode{x03BB} {x}_e\unicode{x03BB} {e}_v[\begin{array}{c}{\mathrm{song}}^{\star }(x)\hskip0.35em \wedge \hskip0.35em {\mathrm{sing}}^{\star }(e)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}(e)=3\cdot \mathrm{d}\mathrm{a}\mathrm{y}\mathrm{s}\hskip0.35em \wedge \hskip0.35em \mathrm{\exists}{\mathrm{u}}_{\mathrm{e}}\mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(e,u)\end{array}]. $
	(b)	Verb denotation
		$ [\hskip0.1em [\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}]\hskip0.1em ]=\unicode{x03BB} {x}_e\unicode{x03BB} {e}_v[{\mathrm{sing}}^{\star }(e)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)]. $
	(c)	VP composition by EDR plus elimination of identical sing⋆(e) and Theme(e,x) conjuncts by logical idempotence.
		$ [\hskip0.1em [\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{g}\mathrm{e}]\hskip0.1em ]=\unicode{x03BB} {x}_e\unicode{x03BB} {e}_v[\begin{array}{c}{\mathrm{song}}^{\star }(x)\hskip0.35em \wedge \hskip0.35em {\mathrm{sing}}^{\star }(e)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)\\ {}\hskip0.35em \wedge \hskip0.35em \mathrm{d}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}(e)=3\cdot \mathrm{d}\mathrm{a}\mathrm{y}\mathrm{s}\hskip0.35em \wedge \hskip0.35em \mathrm{\exists}{\mathrm{u}}_{\mathrm{e}}[\mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(e,u)]\end{array}]. $
	(d)	Outermost Existential Closure of the variable x
		$ {\mathrm{EC}}_{\mathrm{outer}}([\hskip0.1em [\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{s}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{t}\mathrm{i}\mathrm{a}\mathrm{n}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{g}\mathrm{e}]\hskip0.1em ]) $
		$ =\unicode{x03BB} {e}_v\exists {x}_e\left[\begin{array}{c}{\mathrm{song}}^{\star }(x)\wedge {\mathrm{sing}}^{\star }(e)\wedge \mathrm{Theme}\left(e,x\right)\\ {}\wedge \mathrm{duration}\left(\mathrm{e}\right)=3\cdot \mathrm{days}\wedge \exists {\mathrm{u}}_{\mathrm{e}}\left[\mathrm{Agent}\left(e,u\right)\right]\end{array}\right]. $
	(e)	Voice and Event Identification (EI) by Kratzer (Reference Kratzer, Rooryck and Zaring1996)
		(i)	$ [\hskip0.1em [\mathrm{V}\mathrm{o}\mathrm{i}\mathrm{c}\mathrm{e}]\hskip0.1em ]=\lambda {y}_e\lambda {e}_v[\mathrm{A}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{t}(e,y)]\hskip0.24em $
		(ii)	Voice-VP composition by Event Identification (EI)
			$ \mathrm{EI}\left({f}_{\left\langle e,\left\langle v,t\right\rangle \right\rangle },{g}_{\left\langle v,t\right\rangle}\right)\to \lambda {y}_e\lambda {e}_v\left[f(y)(e)\wedge g(e)\right] $ (Kratzer Reference Kratzer, Rooryck and Zaring1996: 122)
			$ \mathrm{E}\mathrm{I}([\hskip0.1em [\mathrm{V}\mathrm{o}\mathrm{i}\mathrm{c}\mathrm{e}]\hskip0.1em ],[\hskip0.1em [\mathrm{V}\mathrm{P}]\hskip0.1em ]) $
			$ =\lambda {y}_e\lambda {e}_v\left[\mathrm{Agent}\left(e,y\right)\hskip-3.6em \wedge \exists {x}_e\left[{\mathrm{song}}^{\star }(x)\wedge {\mathrm{sing}}^{\star }(e)\wedge \mathrm{Theme}\left(e,x\right)\wedge \mathrm{duration}(e)\hskip-20.6em =3\cdot \mathrm{days}\wedge \exists {u}_e\;\mathrm{Agent}\left(e,u\right)\right]\right] $
		(iii)	Subject-VP composition
			Subject denotation: $ [\hskip0.1em [\mathrm{t}\mathrm{a}]\hskip0.1em ]=\mathrm{h}\mathrm{e} $
			$ [\hskip0.1em [\mathrm{VoiceP}]\hskip0.1em ]([\hskip0.1em [\mathrm{ta}]\hskip0.1em ])=\lambda {e}_v[\mathrm{Agent}\left(e,\mathrm{he}\right)\hskip-5.6em \wedge \exists {x}_e\left[\begin{array}{c}{\mathrm{song}}^{\star }(x)\wedge {\mathrm{sing}}^{\star }(e)\wedge \mathrm{Theme}\left(e,x\right)\\ {}\wedge \mathrm{duration}(e)=3\cdot \mathrm{days}\wedge \exists {u}_e.\mathrm{Agent}\left(e,u\right)\end{array}\right]]. $
		(iv)	Agent existential simplification
			$ \mathrm{Agent}\left(e,\mathrm{he}\right) $ entails $ \exists {\mathrm{u}}_{\mathrm{e}}\mathrm{Agent}\left(e,u\right) $ , allowing removal of redundant conjuncts.
		(v)	Final Event closure
			$ \exists {e}_v\exists {x}_e\left[\begin{array}{c}\mathrm{Agent}\left(e,\mathrm{he}\right)\wedge {\mathrm{song}}^{\star }(x)\wedge {\mathrm{sing}}^{\star }(e)\wedge \mathrm{Theme}\left(e,x\right)\\ {}\wedge \mathrm{duration}(e)=3\cdot \mathrm{days}\end{array}\right]. $

Setting aside tense and the specific temporal contribution of perfective ‑le, (64) yields the intended interpretation: he sang songs for three days. Although the duration phrase is interpreted strictly within the nominal constituent, its condition nonetheless predicates over the same event variable introduced by the matrix verb. Under this analysis, the construction is narrowly licensed because the verbal predicate must align with the telic event requirements preserved in the duration‑modified nominal relation.

A significant advantage of the EDR analysis is its ability to capture a core restriction on fake durative nominal modification. As observed by Zhang and Liu (Reference Zhang, Liu, Dong, Hong, Lin and Jin2024: 465), nominals modified by a fake duration quantifier cannot serve as specific, definite or quantified objects.Footnote ⁹ Because the individual variable associated with the head noun in these cases is already closed by a demonstrative, universal or existential operator, it is no longer available for further composition, making EDR inapplicable. Ultimately, relying on a qualia-derived relation alongside EDR provides an elegant solution to this syntax-semantics mismatch without complicating the underlying syntax.

7.2. Canonical fake modification via a head-licensed qualia relation and sentence-level Restrict

This section demonstrates that canonical fake modification exemplifies another case in which a qualia‑induced event argument plays a crucial role in semantic composition. As noted by Huang (Reference Huang2008) and Wenshan Li (Reference Li2021), among others, the pre‑predicate DP‑de‑NP in fake modification does not receive a standard possessive interpretation. Instead, it functions to identify participants in the event evaluated by the subsequent V‑de‑Manner predicate. For instance, in (65) (repeated from 7a), while Daiziying de yuqiu ‘Daiziying’s shuttlecock’ serves as the subject of the sentence, Daiziying and yuqiu ‘shuttlecock’ assume the roles of Agent and Theme of the matrix event, respectively.

This structure shares a close verb-copying paraphrase in (66), where the marker de is replaced by the matrix verb da ‘play’.

This structural and interpretive similarity between these two patterns supports an eventive construal of the pre-predicate DP-de-NP, regardless of whether fake-modification and verb-copying are derivationally related. For the current yuqiu example, this dependency is captured by resolving R to an event-dependent TELIC-derived relation provided by the lexical head noun.

Following the cumulative-closure convention adopted in Section 7, the relevant lexical noun and verb denotations are shown in (67).

(67)	Lexically cumulative noun and verb
	(a)	$ [\hskip0.1em [\mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}]\hskip0.1em ]=\lambda {x}_e.{\mathrm{shuttlecock}}^{\star }(x). $
	(b)	$ [\hskip0.1em [\mathrm{d}\mathrm{a}]\hskip0.1em ]=\lambda {x}_e\lambda {e}_v[{\mathrm{play}}^{\star }(e)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)] $

Here yuqiu contributes a standard individual property; the eventive badminton-playing interpretation is not inherent to the nominal denotation but is derived through the event-dependent relation in (69).

Setting aside the internal semantics of postverbal de, which phonologically attaches to the preceding verb, we treat the entire constituent de hen hao ‘DE very well’ as a type-preserving modifier. Because da retains an unsaturated internal argument, the modifier applies pointwise over it, mapping an expression of type ⟨e,⟨v,t⟩⟩ onto another expression of the same type, as shown in (68). This yields the main predicate denotation for (65).

(68)	(a)	$ [\hskip0.1em [\mathrm{d}\mathrm{e}\hskip0.5em \mathrm{h}\mathrm{e}\mathrm{n}\hskip0.5em \mathrm{h}\mathrm{a}\mathrm{o}]\hskip0.1em ]=\lambda {P}_{\langle e,\langle v,t\rangle \rangle}\hskip0.5em \lambda {x}_e\hskip0.2em \lambda {e}_v\hskip0.1em [P(x)(e)\hskip0.35em \wedge \hskip0.35em \mathrm{v}\mathrm{e}\mathrm{r}\mathrm{y}-\mathrm{w}\mathrm{e}\mathrm{l}\mathrm{l}(e)] $
		Type $ \left\langle \left\langle e,\left\langle v,t\right\rangle \right\rangle, \left\langle e,\left\langle v,t\right\rangle \right\rangle \right\rangle $
	(b)	$ [\hskip0.1em [\mathrm{d}\mathrm{a}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{h}\mathrm{e}\mathrm{n}\hskip0.5em \mathrm{h}\mathrm{a}\mathrm{o}]\hskip0.1em ]=\lambda {x}_e\lambda {e}_v[{\mathrm{play}}^{\star }(e)\hskip0.35em \wedge \hskip0.35em \mathrm{T}\mathrm{h}\mathrm{e}\mathrm{m}\mathrm{e}(e,x)\hskip0.35em \wedge \hskip0.35em \mathrm{v}\mathrm{e}\mathrm{r}\mathrm{y}-\mathrm{w}\mathrm{e}\mathrm{l}\mathrm{l}(e)] $

Turning to the subject DP and its semantic composition, selective binding into an event-dependent qualia role can produce a relation that links a role filler, a nominal participant and an event. Applied to yuqiu ‘shuttlecock’, the relevant head-licensed relation is the TELIC performance relation, which serves as a value for the relation variable R introduced by de.

(69)	TELIC-derived relation provided by the head noun yuqiu
	(a)	$ {R}_{\mathrm{yuqiu}}=\mathrm{S}{\mathrm{B}}_{\mathrm{TELIC}}\left(\mathrm{yuqiu}\right)\in \mathrm{SourceR}\left(\mathrm{yuqiu}\right) $
	(b)	$ \mathrm{S}{\mathrm{B}}_{\mathrm{TELIC}}\left(\mathrm{yuqiu}\right)=\lambda {u}_e\lambda {x}_e\lambda {e}_v\left[\begin{array}{c}{\mathrm{shuttlecock}}^{\star }(x)\wedge {\mathrm{play}}^{\star }(e)\\ {}\wedge \mathrm{Agent}\left(e,u\right)\wedge \mathrm{Theme}\left(e,x\right)\end{array}\right] $

Let the possessive DP Daiziying denote the individual d. Applying the linker de to d and resolving R to the TELIC-induced relation in (69) yields the event-dependent nominal relation in (70).

(70)	DP subject composition
	(a)	$ [\hskip0.1em [\mathrm{D}\mathrm{a}\mathrm{i}\mathrm{z}\mathrm{i}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}]\hskip0.1em ]=[\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ](\mathrm{d})(\mathrm{S}{\mathrm{B}}_{\mathrm{TELIC}}(\mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u})) $
	(b)	$ [\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ]=\lambda {\alpha}_e\lambda {R}_{\langle e,\langle e,\langle v,t\rangle \rangle \rangle}\lambda {x}_e.R(\alpha )(x) $
	(c)	$ [\hskip0.1em [\mathrm{D}\mathrm{a}\mathrm{i}\mathrm{z}\mathrm{i}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}]\hskip0.1em ]=[\hskip0.1em [de]\hskip0.1em ](\mathrm{d})=\lambda {R}_{\langle e,\langle e,\langle v,t\rangle \rangle \rangle}\lambda {x}_e.R(\mathrm{d})(x) $
	(d)	$ [\hskip0.1em [\mathrm{d}\mathrm{e}]\hskip0.1em ](\mathrm{d})(\mathrm{S}{\mathrm{B}}_{\mathrm{TELIC}}(\mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}))=\lambda {x}_e.\mathrm{S}{\mathrm{B}}_{\mathrm{TELIC}}(\mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u})(\mathrm{d})(x) $
		$ =\lambda {x}_e\lambda {e}_v\left[\begin{array}{c}{\mathrm{shuttlecock}}^{\star }(x)\wedge {\mathrm{play}}^{\star }(e)\\ {}\wedge \mathrm{Agent}\left(e,\mathrm{d}\right)\wedge \mathrm{Theme}\left(e,x\right)\end{array}\right] $

The resulting denotation of the subject DP in (70) is of type ⟨e,⟨v,t⟩⟩, establishing a relation over the Theme argument x and event argument e, where the agent is specified as d.

This subject DP denotation must now combine with the predicate denotation in (68b). As shown in (68b), the main predicate is also of the same type ⟨e,⟨v,t⟩⟩. Functional Application therefore cannot apply, since neither takes the other as an argument. They can, however, combine via Event-dependent Restrict (EDR, defined in (63)), which identifies their respective nominal and event arguments. Applying EDR to (71a) initially produces duplicate occurrences of play^⋆(e) and Theme(e,x) – one from (68b) and one from (70d). These duplicates are eliminated by logical idempotence, resulting in (71b).

(71)	Subject-predicate composition via Event-dependent Restrict (EDR)
	(a)	$ [\hskip0.1em [\mathrm{D}\mathrm{a}\mathrm{i}\mathrm{z}\mathrm{i}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}\hskip0.5em \mathrm{d}\mathrm{a}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{h}\mathrm{e}\mathrm{n}\hskip0.5em \mathrm{h}\mathrm{a}\mathrm{o}]\hskip0.1em ] $
		$ =\mathrm{E}\mathrm{D}\mathrm{R}([\hskip0.1em [\mathrm{D}\mathrm{a}\mathrm{i}\mathrm{z}\mathrm{i}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}]\hskip0.1em ],[\hskip0.1em [\mathrm{d}\mathrm{a}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{h}\mathrm{e}\mathrm{n}\hskip0.5em \mathrm{h}\mathrm{a}\mathrm{o}]\hskip0.1em ]) $
	(b)	$ [\hskip0.1em [\mathrm{D}\mathrm{a}\mathrm{i}\mathrm{z}\mathrm{i}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}\hskip0.5em \mathrm{d}\mathrm{a}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{h}\mathrm{e}\mathrm{n}\hskip0.5em \mathrm{h}\mathrm{a}\mathrm{o}]\hskip0.1em ] $
		$ =\lambda {x}_e\lambda {e}_v\left[\begin{array}{c}{\mathrm{shuttlecock}}^{\star }(x)\wedge {\mathrm{play}}^{\star }(e)\\ {}\wedge \mathrm{Agent}\left(e,\mathrm{d}\right)\wedge \mathrm{Theme}\left(e,x\right)\wedge \mathrm{very}-\mathrm{well}(e)\end{array}\right]. $

Notably, the above semantic composition lacks a separate Voice-composition step. This is because Daiziying de yuqiu ‘Daiziying’s shuttlecock’ itself constitutes the subject and the pre-de individual Daiziying has already saturated the performer argument of the TELIC-derived relation in (69). Consequently, the Agent condition in (71b) is contributed by the pre-predicate nominal relation itself and no independent Voice projection is required.

A desirable consequence of the EDR analysis is that the event predicate contributed by the TELIC-derived relation need not be lexically identical to the predicate contributed by the main verb. For example, in (72), the most plausible TELIC-associated event for gāngqín ‘piano’ is playing, whereas the main predicate lian ‘practice’ or biaoyan ‘perform’ characterizes events of practicing or performing, respectively.

Although the playing and practicing/performing predicates differ lexically, they can both describe the same event: a specific playing event may simultaneously be a practicing or performing event. In this case, there are no duplicate occurrences of the predicate play^⋆(e). Instead, the truth conditions contain one play^⋆(e) condition and one perform^⋆(e) condition, where the event denoted by the variable e is simultaneously a practicing or performing event. Consequently, EDR requires the TELIC-derived and main-predicate conditions to hold of the same event and nominal participant, but it does not demand lexical identity. In this respect, our analysis departs from strict verb-copying accounts of fake modification, which derive the construction from an underlying structure containing two identical copies of the verb.

While EDR identifies the corresponding individual and event arguments of the pre-predicate nominal relation and the main predicate, it leaves the shared individual argument unsaturated. Since no subsequent expression saturates this argument, sentence-level EC applies to bind x, yielding an event predicate of type ⟨v,t⟩.

(73)	Outermost Existential Closure
		$ {\mathrm{EC}}_{\mathrm{outer}}(\mathrm{E}\mathrm{D}\mathrm{R}([\hskip0.1em [\mathrm{D}\mathrm{a}\mathrm{i}\mathrm{z}\mathrm{i}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{g}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{y}\mathrm{u}\mathrm{q}\mathrm{i}\mathrm{u}]\hskip0.1em ],[\hskip0.1em [\mathrm{d}\mathrm{a}\hskip0.5em \mathrm{d}\mathrm{e}\hskip0.5em \mathrm{h}\mathrm{e}\mathrm{n}\hskip0.5em \mathrm{h}\mathrm{a}\mathrm{o}]\hskip0.1em ])) $
		$ =\lambda {e}_v\exists {x}_e\left[\begin{array}{c}{\mathrm{shuttlecock}}^{\star }(x)\wedge {\mathrm{play}}^{\star }(e)\wedge \mathrm{Agent}\left(e,d\right)\\ {}\wedge \mathrm{Theme}\left(e,x\right)\wedge \mathrm{very}-\mathrm{well}(e)\end{array}\right] $

In an episodic context, an independently available existential operator, possibly a covert aspectual operator, existentially closes the event variable, yielding the truth conditions in (74).

(74)	Episodic event closure
	$ \exists {e}_v\exists {x}_e\left[\begin{array}{c}{\mathrm{shuttlecock}}^{\star }(x)\wedge {\mathrm{play}}^{\star }(e)\wedge \mathrm{Agent}\left(e,\mathrm{d}\right)\\ {}\wedge \mathrm{Theme}\left(e,x\right)\wedge \mathrm{very}-\mathrm{well}(e)\end{array}\right] $

In (74), since play^⋆ and shuttlecock^⋆ are cumulative and their participant relations summative, they yield atelic or unbounded interpretations, thereby supporting activity readings characteristic of the fake modification construction (cf. Krifka Reference Krifka, Sag and Szabolcsi1992, Reference Krifka and Rothstein1998). In generic contexts, by contrast, the Gen operator yields the generic reading. We will not pursue the details of this interpretation here.

The above derivations illustrate how EDR operates by combining two already well-typed event-dependent relations, while EC simply closes the remaining unsaturated individual argument. The mechanisms used here are closely aligned with those employed for duration pseudo-modification. In particular, both event-sensitive constructions use the same TELIC-induced relation to feed non-saturating EDR, followed by EC. They differ only in how they construct the event-dependent nominal relation. Duration pseudo-modification derives the relation with a higher-order duration modifier that applies to an EC-derived telic event profile before EDR. In canonical fake modification, the pre-de DP first saturates the role filler (i.e. the Agent) of the qualia-induced TELIC predicate, thereby producing a relation over individuals and events. The shared mechanisms of these two constructions provide strong support for a qualia-dependent approach to a unified semantic linker analysis of de.