Angelopoulos 2026: On clausal complementation, once more @cite{angelopoulos-2026} #
Nikos Angelopoulos (2026). Natural Language & Linguistic Theory 44:26. DOI 10.1007/s11049-026-09711-w.
Three puzzles, one mechanism #
The paper resolves three puzzles about Modern Greek oti (content complementizer) and pu (situation/factive complementizer):
- Internal/external argument asymmetry. Both oti- and pu-clauses pattern as internal arguments under three diagnostics — clitic doubling, passivization, A-bar extraction transparency — yet both are banned from the external argument position.
- Near-complementary distribution. oti combines with content-selecting matrix verbs (ipe 'said', pistévo 'believe', xeri 'know'); pu combines with situation- selecting emotive factives (metanjose 'regret', aresi 'like / appeal to', xérome 'be happy').
- Novel stativity restriction on complement pu. Complement pu-clauses require a stative matrix predicate. The restriction vanishes in adjunct, relative, and interrogative pu uses.
The unifying mechanism: oti and pu bear an uninterpretable
[n]-feature checked by a light noun in their specifier
(@cite{arsenijevic-2009}, @cite{moltmann-2019}). The light noun is
licensed by Hale-Keyser noun-incorporation (@cite{hale-keyser-1993})
into a lexical host (v_State or v_Event); incorporation into a
functional head (T) is impossible. Aspectual head v_State
selects either content- or situation-typed [n]; v_Event selects
only content-typed [n]. Adjuncts select-upward
(@cite{bruening-2013}, @cite{hewett-2023}, @cite{hunter-2015},
@cite{neeleman-philip-tanaka-vandekoot-2023}), exempting non-
complement pu from v_State.
Hedges (per audit) #
- Stativity restriction credit. Noted in @cite{roussou-2019} (Angelopoulos credits her example as the source of his ex. 22b discriminating the stativity contrast); the systematic generalization across complement pu-clauses is presented as novel in @cite{angelopoulos-2026}, derived here from v_State c-selection.
- Manner-adverb diagnostic. Unidirectional in the literature. @cite{maienborn-2005} shows Kimian states allow some manner- like modification; the converse (allows manner ⟹ eventive) is contested.
- Clitic doubling as argumenthood. Exposed as a construction
fact (
doubledByClitic), NOT axiomatized as biconditional withisArgument. @cite{anagnostopoulou-2003} ties doubling to specificity, not bare argumenthood. - Crosslinguistic scope. Greek with cross-references to @cite{major-2024} (Uyghur dep) and @cite{bochnak-hanink-2021} (Washo). Two-language extension, not a typology. Korean -ko, Japanese to/koto, Romance que/de, Hebrew še/ki avoided.
Cross-framework engagement #
- Theoretical ally: @cite{bruening-2025} ("C-Selection
Irreducibility") makes a parallel argument for bidirectional
feature-driven selection in coordination, formalised in
Phenomena/Coordination/Studies/Bruening2025.lean. The bidirectional-selection move here generalises Bruening's argument-side claim to the adjunct domain. - Silent divergence vs HPSG: HPSG selection is intrinsically
head-driven (the SUBCAT/COMPS list lives on the head); the
paper's adjunct-selects-host machinery has no HPSG counterpart
yet formalised in linglib. A future
Theories/Syntax/HPSG/Complementation.leanwould let us state the cross-framework refutation directly. - Silent divergence vs CCG: CCG adjuncts are
X/Xmodifiers that don't probe their host; the bidirectional-selection move is also silently divergent from CCG. - Encoding gap: the §6 refutation theorems are conditional
(premises about constructive
SyntacticObjectwitnesses); the substantive instantiation requires aStep.applyAdjunctCheckeddual toSelection.lean's checked Merge, which is queued substrate. As written, both refutation theorems function as paper-attributed promises rather than constructive refutations.
Studies-local apparatus (per audit) #
vAspectHeadenum (vState / vEvent): per the @cite{borer-2005} / @cite{merchant-2019} light-v split. Not promoted to substrate yet — single paper-anchored consumer; promote toTheories/Syntax/Minimalist/LightV.leanwhen ≥ 2 consumers.adjunctSelectspredicate: bidirectional selection per @cite{bruening-2013} / @cite{hewett-2023} / @cite{hunter-2015} / @cite{neeleman-philip-tanaka-vandekoot-2023}. Studies-local; promote toTheories/Syntax/Minimalist/AdjunctSelection.leanwhen ≥ 2 consumers.hkIncorporablepredicate: @cite{hale-keyser-1993} noun- incorporation. Distinct fromPhase.leanD-incorporation (Davies-Dubinsky 2003 phase deactivation) andTypology.Voicenoun-incorporation (Beavers-Udayana 2022 type-shifting taxonomy). The three are coincidentally named but structurally distinct; do NOT unify into a singleIncorporation.leansubstrate.
The light-v aspectual head split (@cite{borer-2005}, @cite{merchant-2019}). v_State introduces a stative situation argument; v_Event introduces an eventive one. These select different sorts of [n]-licensee in the spec of their CP complement.
Studies-local. Promote to substrate when a second paper- anchored consumer (e.g., a future Borer 2005 or Merchant 2019 study file) consumes it.
- vState : vAspectHead
- vEvent : vAspectHead
Instances For
Equations
- Phenomena.Complementation.Studies.Angelopoulos2026.instDecidableEqVAspectHead x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
Which sort of [n]-licensee an aspectual head accepts:
v_State accepts both content and situation; v_Event accepts
only content (paper §4.1). This asymmetry is the encoded
table; the underlying derivation from a deeper aspectual-
stativity feature (Borer 2005: situations are stative
aspectual objects, dynamic v cannot select stative complement)
is queued as a TODO. As written, this def-as-table is
paper-fidelity, not a structural derivation.
Prop-valued per CLAUDE.md (no Bool for predicate-shape
data); decidability instance below.
Equations
- Phenomena.Complementation.Studies.Angelopoulos2026.vAspectHead.vState.acceptsSort x✝ = True
- Phenomena.Complementation.Studies.Angelopoulos2026.vAspectHead.vEvent.acceptsSort Phenomena.Complementation.Studies.Bondarenko2022.NominalSort.content = True
- Phenomena.Complementation.Studies.Angelopoulos2026.vAspectHead.vEvent.acceptsSort Phenomena.Complementation.Studies.Bondarenko2022.NominalSort.situation = False
Instances For
Equations
- One or more equations did not get rendered due to their size.
@cite{hale-keyser-1993} noun-incorporation predicate: a light-noun [n] can incorporate into a category iff that category is lexical (in the @cite{panagiotidis-2015} sense: bears interpretable categorial features, not uninterpretable copies). Functional heads (T, C, D) cannot host incorporation.
Equations
- Phenomena.Complementation.Studies.Angelopoulos2026.hkIncorporable host = (host = Minimalist.Cat.v ∨ host = Minimalist.Cat.V ∨ host = Minimalist.Cat.n ∨ host = Minimalist.Cat.N)
Instances For
Equations
Adjunct-selection predicate: an adjunct selects its host
rather than being selected by it (@cite{bruening-2013},
@cite{hewett-2023}, @cite{hunter-2015},
@cite{neeleman-philip-tanaka-vandekoot-2023}). Studies-local;
a coarse abstraction over the actual checked-Merge dual that
a future AdjunctSelection.lean substrate would expose.
Equations
Instances For
Was the matrix verb's complement clitic-doubled? Construction fact, NOT axiomatized to imply argumenthood. Cf. fn 4 of the paper acknowledging @cite{angelopoulos-michelioudakis-2023} on doubling-as-Agree.
Equations
Instances For
The bare oti-clause yields the explanans reading: "Maria explained well that the Earth is round" — the embedded clause IS the explanation (paper ex. 4a).
- bareOti : ExplanansReading
Instances For
Equations
- Phenomena.Complementation.Studies.Angelopoulos2026.instDecidableEqExplanansReading x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
The nominalized to oti-clause yields the explanandum reading: "Maria explained well the fact that the Earth is round" — the embedded clause is the fact being explained (paper ex. 4c).
- toOti : ExplanandumReading
Instances For
Equations
- Phenomena.Complementation.Studies.Angelopoulos2026.instDecidableEqExplanandumReading x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
The oti C-head: bears [uN] (= the [n]-feature in the paper's notation), selecting a light noun in its specifier.
Equations
Instances For
The pu C-head: same selectional contract as oti — bears the [n]-feature. The content/situation distinction surfaces not at C but at the matrix v (§5).
Equations
Instances For
The light-noun-in-Spec configuration: nullN from
Selection.lean merged into the spec of an oti/pu C head
via specifier Merge (Step.applyChecked .emL).
The C head retains its [n] in pending after emL (specifier
Merge does not consume features in the c-selection system; cf.
applyChecked_emL). The [n] is checked when the resulting CP
is merged as a complement of v whose own selectional features
require a CP whose Spec contains [n].
Equations
- Phenomena.Complementation.Studies.Angelopoulos2026.lightNounInSpec cHead id = (Minimalist.nullN id).node cHead
Instances For
Functional heads (T, C, D) bear uninterpretable categorial feature copies per @cite{panagiotidis-2015}; they cannot host H&K incorporation.
C is functional; cannot host incorporation.
D is functional; cannot host incorporation.
v is lexical (a categorizer per @cite{panagiotidis-2015}); hosts incorporation. This is the case Angelopoulos exploits: a light noun in Spec,CP can incorporate into a higher v_State or v_Event, satisfying its licensing requirement.
V is lexical; hosts incorporation.
The three functional-head cells are jointly non-incorporation-
hosts. This is the substrate-level fact underlying Puzzle 1
(in/out asymmetry); the geometric step "from Spec,TP, the
closest c-commander is T, and incorporation cannot reach
lexical v without violating Sportiche-2005 locality" is
asserted in prose here and queued for formalisation when a
head-movement substrate (Step.applyHeadMove) lands.
Paper ex. 11, 12: bare oti/pu-clauses are ungrammatical in the external argument position.
The internal-argument cases work because the light noun's closest lexical host (v_State or v_Event) is c-commanding the CP from above its complement position.
v_State accepts both content and situation [n]-typed light nouns.
v_Event accepts content but rejects situation.
Puzzle 3 (stativity restriction, derived from the table). A pu-clause bears situation-typed [n]; v_Event rejects it; therefore pu cannot complement an eventive matrix predicate.
Paper ex. 23c: *θimose mésa se péde leptá pu psifístike o nómos 'got angry within five minutes that the law was voted'.
The in-adverbial forces eventive interpretation; eventive →
v_Event; v_Event ↛ situation; pu blocked.
TODO: derive acceptsSort from a Borer-2005 stativity feature
on the head and a stativity feature on the licensee, rather
than table-stipulating it. The current encoding is the
paper-fidelity stub form; a true derivation is queued.
oti (content-typed [n]) is licensed under either v.
Predicts paper ex. 21b: arketos kosmos δen siniδitopii efkola oti i periγrafiki kanones ine δjaforetiki... — sinidhitopió
'realize' takes a manner adverb (eventive ⟹ v_Event), and
oti is fine.
Refutation against @cite{bondarenko-2022}'s transparent
Syntax-Semantics mapping — substantive form, consuming the
Chapter 4 type-theoretic predicates from
Phenomena.Complementation.Studies.Bondarenko2022.
The substantive content: Greek bare oti in the explanans
reading (paper ex. 4a) instantiates the (semantic-)bare clause
type composing via PM with v's situation argument. Bondarenko
correctly predicts this composition (bareCP_composes_via_PM).
But the same bare oti passes argument-position diagnostics
(clitic doubling — paper ex. 3a — and passivization — ex. 6a).
Bondarenko's prediction bareCP_satisfies_no_theta says no
Θ-head saturates a bare CP. Therefore: either (i) the diagnostics
track something other than Θ-saturation, or (ii) the prediction
fails. Angelopoulos chooses (i) — autonomy of syntax — which IS
Bondarenko's transparent S-S mapping rejected.
Stated as the conjunction witnessing the autonomy claim: bare CP composes via PM AND no Θ-head saturates it. Angelopoulos's diagnostics force the position↔Θ-saturation correlation that Bondarenko's transparent mapping presupposes.
The original conditional refutation, preserved for the
paper-fidelity transparentSSMapping def-as-table form. The
substantive refutation is
angelopoulos_explanans_breaks_position_theta_correlation above.
Refutation against @cite{bochnak-hanink-2021} / Deal 2026 "selection limited to argument clauses". If selection were limited to argument clauses, then a pu-adjunct clause would not be selected by anything. But Greek adjunct pu-clauses select their hosts (§4.2 paper, following @cite{hewett-2023}, @cite{hunter-2015}). Witness: paper ex. 24a — Ton timorisan γrigora pu iče afti ti siberifora 'They punished him fast because of the fact that he had this behavior'. The pu- adjunct selects the verbal host; selection is bidirectional.
TODO: instantiate puAdjunctClause constructively from
paper ex. 24a + adjunctSelects machinery. Requires checked-
Merge dual for adjunct-up selection (Step.applyAdjunctChecked),
which is genuinely new substrate not yet in Selection.lean.
The content/situation distinction (Bondarenko 2022) cuts
orthogonally to the doxastic/preferential partition exposed
in Features.Attitudes.Attitude.
The substantive claim is that there is a (preferential,
situation-selecting) verb (e.g., metanjose 'regret') and a
(doxastic, content-selecting) verb (e.g., pistévo 'believe').
However, content/situation selection is not currently a
Fragment field — per integration auditor N3, it is paper-
specific apparatus that lives in this Studies file rather than
in VerbCore. Without a Fragment field, the orthogonality
theorem can only state a Vendler-stativity witness, which is
a proxy for situation-selection (since pu-complements
require stative matrix predicates per Puzzle 3) but not
identical to it.
The honest theorem: there exists a (preferential, stative)
verb. The cross-classification with content/situation
selection requires a Fragment-level selectsClauseSort field,
queued for a follow-up substrate refactor.
And a (doxastic, stative) verb — pistévo 'believe'. The
pair witnesses that attitude class and stativity cross-
classify; together with the prose claim that metanjose
selects pu (situation) and pistévo selects oti
(content), this is the closest we can come to the orthogonality
claim without a selectsClauseSort Fragment field.
The oti-explanans reading uses the content-CP via PM.
The CP oti i Ji ine strojili 'that the Earth is round'
composes with the verb's situation argument by Predicate
Modification, modifying the event of explaining with the
propositional content. Reference: compC from
ClauseDenotation/Content.lean.
The to oti-explanandum reading uses the silent FACT noun + D layer per @cite{moulton-2020}. Composes via Functional Application to deliver an individual argument (the fact being explained).
Reference: existsContentNounCP from ClauseDenotation/Content.
The silent noun is a ContentNoun W whose denotation is
factive (presupposes the proposition true at the actual
world).
Symmetric for pu: the situation-clause yields its denotation
via compPu from ClauseDenotation/Situation.lean.