Chan & Shen (2026): Conditions on wh-the-hell licensing

@cite{chan-shen-2026} @cite{pesetsky-1987} @cite{chou-2012} @cite{merchant-2002} @cite{sato-ngui-2017} @cite{rawlins-2008} @cite{martin-2020} @cite{ippolito-2024} @cite{huang-ochi-2004} @cite{linebarger-1987} @cite{hoeksema-napoli-2008}

Linguistic Inquiry. Advance publication. https://doi.org/10.1162/LING.a.562

Empirical Contribution

Acceptability judgment experiment (N=32 Singlish speakers, two crossed 2×2 factorial designs sharing the Long baseline). Establishes that in Singlish single wh-questions:

• In-situ wh-the-hell is unacceptable: superadditive interaction (WhType × Strategy p < 0.001; DD = 1.15)
• Partial-movement wh-the-hell is acceptable: additive costs only (no interaction p = 0.882; DD = -0.02)
• The same in-situ ban holds for subject wh-in-situ (paper §3.3 ex 22b) — a separate prediction failure for the intervention account (paper §3.4.1, p. 23–24): no higher wh, no intervener, but still bad.

The original observation that in-situ wh-the-hell is bad goes back to @cite{pesetsky-1987} (introducing "aggressively non-D-linked").

Theoretical Contribution

Two-component analysis:

1. POV licensing (@cite{chou-2012}, building on @cite{huang-ochi-2004}): the-hell bears an unvalued POV feature [ud] that must be checked in a Spec-head relation with a POV operator in matrix CP, ascribing the negative attitude of the-hell to the speaker of the utterance.

2. Parasitic movement (@cite{merchant-2002}): the-hell is a modifier adjoined to the wh-head. It cannot move independently — its movement to Spec-CP is parasitic on the wh-phrase's movement.

Predicts: the-hell is licensed iff the wh-phrase reaches matrix Spec-CP. Full and partial movement satisfy this; unselective binding (in-situ) does not.

Comparison with Alternative Accounts

• @cite{den-dikken-giannakidou-2002} (intervention/PI): correctly predicts in-situ bad in multiple wh-questions (intervener present) but wrongly predicts in-situ OK in single wh-questions (no intervener, per @cite{linebarger-1987}'s immediate-scope condition); doubly wrong for subject in-situ where Q is in immediate scope.
• @cite{vu-lohiniva-2020} (AttP, building on @cite{huang-ochi-2004}): correctly predicts full OK and in-situ bad, but wrongly predicts partial bad (paper §3.4.2 ex 32: cannot generate the correct word order with the-hell in matrix Spec-AttP and wh-phrase in embedded Spec-CP).

Cross-linguistic generalization

The single typological parameter @cite{chan-shen-2026} isolate is the modifier's movement profile (ANDLMovementType.parasitic for English/Singlish the-hell vs .independent for Mandarin daodi, @cite{chou-2012}). Other ANDL items — the heck, the fuck, the dickens, in the world, in God's name (@cite{hoeksema-napoli-2008}, @cite{jackendoff-audring-2020}; paper footnote 6) — are predicted to behave like the-hell.

Architecture

Theory-neutral lexical entries (theHell, daodi) live in the respective Fragment files. The Minimalist analysis (POV features, Agree, the licensing predicate) lives in Theories/Syntax/Minimalism/ Core/ANDL.lean. The empirical 2×2 design uses Paradigms/AcceptabilityJudgment.lean. This study file only carries the paper's specific data (six conditions, two DD scores) and the bridge theorems connecting theory to data.

Minimalist POV-feature analysis (formerly Core/ANDL.lean)

The Minimalist (POV-feature) analysis of aggressively non-D-linked (ANDL) wh-modifiers, due to @cite{chou-2012} (building on @cite{huang-ochi-2004}, @cite{merchant-2002}). The theory-neutral lexical entry lives in Core/Lexical/ExpressiveModifier.lean; this section adds the framework-specific syntactic apparatus: an unvalued POV feature [ud] on the modifier, a valued [+d] POV operator merged in matrix C, and Spec-head Agree as the licensing relation.

1. ANDL modifier (e.g., the-hell) carries an unvalued POV feature [ud]: a probe needing valuation.
2. Matrix C carries a valued POV operator [+d]: a goal.
3. Feature checking happens in Spec-head configuration in matrix CP.
4. Therefore the modifier must reach matrix Spec-CP. For parasitic modifiers (English/Singlish the-hell), this requires the wh-host to reach matrix Spec-CP. For independent modifiers (Mandarin daodi), the modifier moves on its own.

def Minimalist.ANDL.povUnvaluedFeature : GramFeature

The unvalued POV feature [ud] borne by ANDL modifiers (@cite{chou-2012}). A probe seeking a [+d] goal in a Spec-head relation.

Equations

• Minimalist.ANDL.povUnvaluedFeature = Minimalist.GramFeature.unvalued (Minimalist.FeatureVal.pov true)

def Minimalist.ANDL.povOperatorFeature : GramFeature

The valued POV feature [+d] on the matrix-C POV operator. The goal that values [ud].

Equations

• Minimalist.ANDL.povOperatorFeature = Minimalist.GramFeature.valued (Minimalist.FeatureVal.pov true)

theorem Minimalist.ANDL.pov_probe_goal_match : featuresMatch povUnvaluedFeature povOperatorFeature = true

The probe-goal pair matches under featuresMatch: same feature type, opposite valuation status — the prerequisite for Agree.

theorem Minimalist.ANDL.andl_pov_unvalued : povUnvaluedFeature.isUnvalued = true

The ANDL modifier's POV feature is unvalued (a probe).

theorem Minimalist.ANDL.pov_operator_valued : povOperatorFeature.isValued = true

The matrix-C POV operator carries a valued feature (a goal).

@[reducible, inline]

abbrev Minimalist.ANDL.LicensedMinimalist (m : Typology.ExpressiveModifier.ExpressiveWhModifier) (whHostReachesMatrixSpecCP : Prop) : Prop

Minimalist licensing: an ANDL modifier is licensed iff a configuration obtains in which povUnvaluedFeature checks against povOperatorFeature in matrix Spec-CP. Operationally:

• For a parasitic modifier, the wh-host must reach matrix Spec-CP (so that the adjoined modifier reaches Spec-CP with it).
• For an independent modifier, the modifier moves to matrix Spec-CP on its own — host reachability is irrelevant.

This is the Minimalist instantiation of the theory-neutral Typology.ExpressiveModifier.Licensed. The Minimalist version doesn't add a separate condition — it identifies "modifier reaches Spec-CP" as the structural realization of "scope position reached".

Equations

• Minimalist.ANDL.LicensedMinimalist m whHostReachesMatrixSpecCP = Typology.ExpressiveModifier.Licensed m whHostReachesMatrixSpecCP

def ChanShen2026.TheHellLicensed (s : Fragments.Singlish.Questions.WhStrategy) : Prop

The-hell is licensed under strategy s iff the Minimalist LicensedMinimalist predicate holds with the wh-host's matrix Spec-CP reachability as the input. For parasitic the-hell, this reduces to "wh-host reaches matrix Spec-CP" — the licensing condition IS the reachability condition.

Equations

• ChanShen2026.TheHellLicensed s = Minimalist.ANDL.LicensedMinimalist Fragments.Singlish.Questions.theHell s.ReachesMatrixSpecCP

@[implicit_reducible]

instance ChanShen2026.instDecidableTheHellLicensed (s : Fragments.Singlish.Questions.WhStrategy) : Decidable (TheHellLicensed s)

Equations

• ChanShen2026.instDecidableTheHellLicensed s = id inferInstance

theorem ChanShen2026.theHellLicensed_iff_reachesSpecCP (s : Fragments.Singlish.Questions.WhStrategy) : TheHellLicensed s ↔ s.ReachesMatrixSpecCP

For parasitic the-hell, licensing reduces to host reachability.

theorem ChanShen2026.fullMovement_licenses_theHell : TheHellLicensed Fragments.Singlish.Questions.fullMovement

Full wh-movement licenses the-hell.

theorem ChanShen2026.partialMovement_licenses_theHell : TheHellLicensed Fragments.Singlish.Questions.partialMovement

Partial wh-movement licenses the-hell.

theorem ChanShen2026.inSitu_blocks_theHell : ¬TheHellLicensed Fragments.Singlish.Questions.whInSitu

Wh-in-situ blocks the-hell.

@[reducible, inline]

abbrev ChanShen2026.Condition : Type

A wh-the-hell condition is a FactorialCondition with two factors: WhType (does the sentence contain the hell?) and the wh-strategy.

Equations

• ChanShen2026.Condition = Paradigms.AcceptabilityJudgment.FactorialCondition Bool Fragments.Singlish.Questions.WhStrategy

def ChanShen2026.whLong : Condition

In-situ comparison conditions (paper §2.1, ex 4):

Equations

• ChanShen2026.whLong = { label := "Wh-Long", level1 := false, level2 := Fragments.Singlish.Questions.fullMovement, sentence := "What you think Natalie is baking at 3am ah?" }

def ChanShen2026.whHellLong : Condition

Equations

• ChanShen2026.whHellLong = { label := "WhHell-Long", level1 := true, level2 := Fragments.Singlish.Questions.fullMovement, sentence := "What the hell you think Natalie is baking at 3am ah?" }

def ChanShen2026.whSitu : Condition

Equations

• ChanShen2026.whSitu = { label := "Wh-Situ", level1 := false, level2 := Fragments.Singlish.Questions.whInSitu, sentence := "You think Natalie is baking what at 3am ah?" }

def ChanShen2026.whHellSitu : Condition

Equations

• ChanShen2026.whHellSitu = { label := "WhHell-Situ", level1 := true, level2 := Fragments.Singlish.Questions.whInSitu, sentence := "You think Natalie is baking what the hell at 3am ah?" }

def ChanShen2026.whPartial : Condition

Partial movement comparison conditions (paper §2.1, ex 6):

Equations

• ChanShen2026.whPartial = { label := "Wh-Partial", level1 := false, level2 := Fragments.Singlish.Questions.partialMovement, sentence := "You think what Natalie is baking at 3am ah?" }

def ChanShen2026.whHellPartial : Condition

Equations

• One or more equations did not get rendered due to their size.

def ChanShen2026.whHellSituSubject : Condition

Subject wh-in-situ comparison (paper §3.3, ex 22). Subject in-situ wh-the-hell is also unacceptable, despite no intervener (single wh-question, Q in immediate scope) — a separate prediction failure for the intervention account.

Equations

• One or more equations did not get rendered due to their size.

def ChanShen2026.insituDD : Paradigms.AcceptabilityJudgment.DDResult

In-situ comparison DD score: large positive (1.15), significant interaction → superadditive penalty for the-hell in-situ.

Equations

• ChanShen2026.insituDD = { comparison := "in-situ vs full movement", dd := 23 / 20, interactionSignificant := true }

def ChanShen2026.partialDD : Paradigms.AcceptabilityJudgment.DDResult

Partial-movement comparison DD score: ≈ 0 (-0.02), no significant interaction → costs are linearly additive.

Equations

• ChanShen2026.partialDD = { comparison := "partial vs full movement", dd := -1 / 50, interactionSignificant := false }

theorem ChanShen2026.insituDD_superadditive : insituDD.Superadditive

The in-situ DD is genuinely positive (superadditive).

theorem ChanShen2026.partialDD_not_superadditive : ¬partialDD.Superadditive

The partial-movement DD is non-positive (additive or below).

theorem ChanShen2026.whHellLong_licensed : TheHellLicensed whHellLong.level2

For each wh-the-hell condition, the strategy's licensing prediction matches the experimental outcome. These theorems break if a condition's strategy changes or if the licensing predicate is redefined — they tie experimental data to theory.

theorem ChanShen2026.whHellPartial_licensed : TheHellLicensed whHellPartial.level2

theorem ChanShen2026.whHellSitu_unlicensed : ¬TheHellLicensed whHellSitu.level2

theorem ChanShen2026.whHellSituSubject_unlicensed : ¬TheHellLicensed whHellSituSubject.level2

def ChanShen2026.empiricalPattern : Paradigms.AcceptabilityJudgment.AccountPredictions 4

The empirical pattern: full ✓, partial ✓, in-situ ✗, subject in-situ ✗.

Equations

• ChanShen2026.empiricalPattern = Paradigms.AcceptabilityJudgment.AccountPredictions.of2x2 True True False False

def ChanShen2026.negativeAttitudeAscription : Paradigms.AcceptabilityJudgment.AccountPredictions 4

Chan & Shen 2026 (negative attitude ascription via POV). The predictions are derived from TheHellLicensed, which derives from WhStrategy.ReachesMatrixSpecCP, which derives from the strategy's WhInterpMechanism.

Equations

• One or more equations did not get rendered due to their size.

def ChanShen2026.denDikkenGiannakidou : Paradigms.AcceptabilityJudgment.AccountPredictions 4

Den Dikken & Giannakidou (2002) intervention account predictions. Their empirical claim is that wh-the-hell is licensed iff Q is in the modifier's immediate scope (@cite{linebarger-1987}). In single wh-questions there is no other wh-phrase to intervene; their account therefore predicts all four single-wh cells acceptable — including object in-situ and subject in-situ, both wrongly. (Paper §3.4.1.)

Equations

• ChanShen2026.denDikkenGiannakidou = Paradigms.AcceptabilityJudgment.AccountPredictions.of2x2 True True True True

def ChanShen2026.vuLohiniva : Paradigms.AcceptabilityJudgment.AccountPredictions 4

Vu & Lohiniva (2020) AttP account predictions. The-hell is base-generated in matrix Spec-AttP; the nearest wh-phrase moves to Spec-AttP to check [+wh] before wh-the-hell moves to Spec-CP.

• Full movement: wh-phrase moves all the way; ✓.
• Partial movement: wh-phrase stops in embedded Spec-CP; the hell is in matrix Spec-AttP. There is no derivation that places them in a single constituent at Spell-Out (paper §3.4.2 ex 32: "no way to generate the correct word order"). Predicts ✗ — wrongly.
• In-situ: wh-phrase doesn't reach Spec-AttP. ✗.
• Subject in-situ: same. ✗. Three out of four right; partial-movement cell is the failure.

Equations

• ChanShen2026.vuLohiniva = Paradigms.AcceptabilityJudgment.AccountPredictions.of2x2 True False False False

theorem ChanShen2026.only_pov_account_matches : negativeAttitudeAscription.Matches empiricalPattern ∧ ¬denDikkenGiannakidou.Matches empiricalPattern ∧ ¬vuLohiniva.Matches empiricalPattern

Only the Chan & Shen (2026) account matches the empirical pattern.

theorem ChanShen2026.insitu_binding_no_pic : ShenHuang2026.constraintsForDependencyType Typology.Question.WhInterpMechanism.unselectiveBinding.toDependencyType = [Phenomena.Islands.IslandSource.semantic]

Singlish wh-in-situ uses binding (not movement), just like Mandarin wh-in-situ in @cite{shen-huang-2026}. Therefore only the Specificity Condition applies — the PIC is inapplicable. This is why Singlish wh-in-situ is island-insensitive (@cite{sato-ngui-2017}: 11b).

Connection: constraintsForDependencyType .binding = [.semantic] (no syntactic / PIC constraint).

theorem ChanShen2026.partial_movement_pic_applies : ShenHuang2026.constraintsForDependencyType Typology.Question.WhInterpMechanism.partialMovement.toDependencyType = ShenHuang2026.constraintsForDependencyType Typology.Question.WhInterpMechanism.overtMovement.toDependencyType

Conversely, partial movement (the second covert step) IS island- sensitive — paper §3.1 ex 15 shows partial movement out of a complex NP is unacceptable. Bridges to Shen & Huang's classification via partialMovement → .movement → [.syntactic, .semantic].

The syntactic POV feature on the-hell is the feature-checking reflex of the semantic PerspectiveP layer (@cite{dayal-2025}). Both encode the requirement that a perspectival center (the speaker, in direct questions) must be identified.

- **Syntactic** (this file): [*ud*] on *the-hell* checked by POV-op
  in matrix C; reaches Spec-CP iff host reaches Spec-CP.
- **Semantic** (`LeftPeriphery.lean`): PerspP introduces PRO with
  `◇¬know(speaker, Ans(Q))` — the possible-ignorance presupposition.

*The-hell*'s negative attitude (speaker finds every possible answer
improbable, @cite{rawlins-2008}; ignorance reading,
@cite{martin-2020}; conventional implicature, @cite{ippolito-2024})
strengthens PerspP's possible-ignorance presupposition. 

def ChanShen2026.theHellSelectionClass : Interfaces.SyntaxSemantics.LeftPeriphery.SelectionClass

Direct wh-the-hell questions select PerspP — they require the speaker as perspectival center (the negative attitude bearer in @cite{chou-2012}'s analysis). Bridges Chan & Shen's syntactic POV apparatus to Dayal's semantic PerspP layer.

Equations

• ChanShen2026.theHellSelectionClass = Interfaces.SyntaxSemantics.LeftPeriphery.SelectionClass.rogativePerspP

theorem ChanShen2026.theHell_no_knowledge : Interfaces.SyntaxSemantics.LeftPeriphery.entailsKnowledge theHellSelectionClass = false

The PerspP-selecting class is precisely the one that does not entail knowledge of the answer — matching the-hell's ignorance component (@cite{martin-2020}). Bridge from LeftPeriphery.

theorem ChanShen2026.theHell_persp_consistent : Interfaces.SyntaxSemantics.LeftPeriphery.perspPConsistent theHellSelectionClass false false = true

The PerspP-selecting class is consistent with the possible-ignorance presupposition (◇¬know(speaker, Ans(Q))) — the semantic side of the-hell's negative attitude (@cite{rawlins-2008}, @cite{ippolito-2024}).

theorem ChanShen2026.theHell_daodi_movement_contrast : Fragments.Singlish.Questions.theHell.movementType = Typology.ExpressiveModifier.ANDLMovementType.parasitic ∧ Fragments.Mandarin.Questions.daodi.movementType = Typology.ExpressiveModifier.ANDLMovementType.independent

The the-hell / daodi minimal pair: same POV feature analysis (povUnvaluedFeature in both); single parametric difference is ANDLMovementType.parasitic vs .independent.

theorem ChanShen2026.daodi_licensed_insitu (P : Prop) : Typology.ExpressiveModifier.Licensed Fragments.Mandarin.Questions.daodi P

Daodi is licensed even with wh-in-situ — it moves independently to matrix Spec-CP. Theory-neutral consequence of the typological parameter, derived via independent_matrix_always_licensed.