@cite{rudin-2025b}: Embedded Intonation and Quotative Complements #
Rudin, Deniz (2025/2026). "Embedded Intonation and Quotative Complements to Verbs of Speech." Linguistic Inquiry, early access. doi:10.1162/ling.a.554.
Empirical Generalizations #
The paper's central observation: verbs of speech systematically split on whether they accept rising-declarative ("quotative") complements:
| Verb | "p" | "p?" | "p" loud | "p" whisp | that p | that wh / Q |
|---|---|---|---|---|---|---|
| say | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ |
| assert | ✓ | ✗ | ✓ | ✓ | ✓ | ✗ |
| yell | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ |
| whisper | ✓ | ✓ | ✗ | ✓ | ✓ | ✗ |
| ask | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ |
Architecture: One Definition, Not Three #
Following mathlib practice, this file has no parallel formalizations.
Felicitous M V cis the single, model-parameterized definition of felicity: a complement is felicitous with a verb predicate iff there exists a witness (event + performance/proposition) with the right ontological properties.IsRudinModel Mis a class with 30 fields, one per cell. This is the statement of @cite{rudin-2025b}'s empirical claim — any model is tested against it.rudinModelis the concreteSpeechVerbs ℕ Bool (FBPerformance Bool) (fbOntology Bool)instantiation — Farkas-Bruce-grounded, with verb predicates defined as the postulate RHS so the meaning postulates hold byrfl.instance : IsRudinModel rudinModeldischarges the 30 cells from the postulates + FB ontology axioms.
There is no separate empirical : Verb → Complement → Felicity
function and no separate predicted decision function. The empirical
matrix and its derivation are the same proposition.
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.instDecidableEqVerb x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
Complement types in the Rudin matrix.
- quoteDecl : Complement
- quoteRising : Complement
- quoteLoud : Complement
- quoteWhispered : Complement
- thatProp : Complement
- thatQuestion : Complement
Instances For
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.instDecidableEqComplement 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
Selector: map a Verb enum to the corresponding model predicate.
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.Verb.toPred M Phenomena.Quotation.Studies.Rudin2025LI.Verb.say = M.SAY
- Phenomena.Quotation.Studies.Rudin2025LI.Verb.toPred M Phenomena.Quotation.Studies.Rudin2025LI.Verb.assert = M.ASSERT
- Phenomena.Quotation.Studies.Rudin2025LI.Verb.toPred M Phenomena.Quotation.Studies.Rudin2025LI.Verb.yell = M.YELL
- Phenomena.Quotation.Studies.Rudin2025LI.Verb.toPred M Phenomena.Quotation.Studies.Rudin2025LI.Verb.whisper = M.WHISPER
- Phenomena.Quotation.Studies.Rudin2025LI.Verb.toPred M Phenomena.Quotation.Studies.Rudin2025LI.Verb.ask = M.ASK
Instances For
A complement is felicitous with a verb predicate in a given model iff there exists a witness — an event-and-performance pair (for quotative complements) or an event-and-content pair (for that-clauses) — satisfying the ontological constraints encoded by the complement type.
Quotative complements constrain the REENACTed performance:
quoteDecl requires a committing linguistic performance,
quoteRising a rising declarative, quoteLoud/quoteWhispered
a committing performance with the marked volume.
Propositional complements constrain the CONTENT denotation:
thatProp requires propositional content, thatQuestion requires
question content.
Verb-class felicity is then derived: a verb's postulates impose constraints on REENACTed performances (or CONTENT denotations); if those constraints conflict with the complement's, no witness exists and the cell is infelicitous.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A SpeechVerbs model satisfies @cite{rudin-2025b}'s empirical
claims about English speech verbs. The 30 fields are exactly the
cells of the verb × complement felicity matrix.
This class IS the empirical claim. There is no separate empirical
function whose values must be reconciled with model predictions —
a model satisfies these facts or it does not.
- say_quoteDecl : Felicitous M M.SAY Complement.quoteDecl
- say_quoteRising : Felicitous M M.SAY Complement.quoteRising
- say_quoteLoud : Felicitous M M.SAY Complement.quoteLoud
- say_quoteWhispered : Felicitous M M.SAY Complement.quoteWhispered
- say_thatProp : Felicitous M M.SAY Complement.thatProp
- say_thatQuestion : ¬Felicitous M M.SAY Complement.thatQuestion
- assert_quoteDecl : Felicitous M M.ASSERT Complement.quoteDecl
- assert_quoteRising : ¬Felicitous M M.ASSERT Complement.quoteRising
- assert_quoteLoud : Felicitous M M.ASSERT Complement.quoteLoud
- assert_quoteWhispered : Felicitous M M.ASSERT Complement.quoteWhispered
- assert_thatProp : Felicitous M M.ASSERT Complement.thatProp
- assert_thatQuestion : ¬Felicitous M M.ASSERT Complement.thatQuestion
- yell_quoteDecl : Felicitous M M.YELL Complement.quoteDecl
- yell_quoteRising : Felicitous M M.YELL Complement.quoteRising
- yell_quoteLoud : Felicitous M M.YELL Complement.quoteLoud
- yell_quoteWhispered : ¬Felicitous M M.YELL Complement.quoteWhispered
- yell_thatProp : Felicitous M M.YELL Complement.thatProp
- yell_thatQuestion : ¬Felicitous M M.YELL Complement.thatQuestion
- whisper_quoteDecl : Felicitous M M.WHISPER Complement.quoteDecl
- whisper_quoteRising : Felicitous M M.WHISPER Complement.quoteRising
- whisper_quoteLoud : ¬Felicitous M M.WHISPER Complement.quoteLoud
- whisper_quoteWhispered : Felicitous M M.WHISPER Complement.quoteWhispered
- whisper_thatProp : Felicitous M M.WHISPER Complement.thatProp
- whisper_thatQuestion : ¬Felicitous M M.WHISPER Complement.thatQuestion
- ask_quoteDecl : ¬Felicitous M M.ASK Complement.quoteDecl
- ask_quoteRising : Felicitous M M.ASK Complement.quoteRising
- ask_quoteLoud : ¬Felicitous M M.ASK Complement.quoteLoud
- ask_quoteWhispered : ¬Felicitous M M.ASK Complement.quoteWhispered
- ask_thatProp : ¬Felicitous M M.ASK Complement.thatProp
- ask_thatQuestion : Felicitous M M.ASK Complement.thatQuestion
Instances
We now build a concrete SpeechVerbs model over FBPerformance Bool
with fbOntology as its performance ontology, and show it satisfies
IsRudinModel. The model uses ℕ as the time type and Bool as the
semantic-object type (true ↦ proposition, false ↦ question).
Verb predicates are defined as the postulate RHS, so the meaning
postulates hold by rfl. The discriminator for verb classes is
runtime.start (0 = SAY, 1 = ASSERT, 2 = YELL, 3 = WHISPER, 4 = ASK).
REENACT and CONTENT are defined per verb class to give the right
witnesses and exclusions.
A canonical event for each verb class, indexed by runtime.start.
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.E n = { runtime := { start := n, finish := n, valid := ⋯ }, sort := Semantics.Events.EventSort.action }
Instances For
The REENACT relation: per verb-class events have different REENACT
targets, chosen so the postulates' universal quantifiers reduce to
obvious tautologies (e.g., for SAY events, REENACT only relates to
LINGMAT performances, so SAY's postulate ∀u, REENACT → LINGMAT
is vacuously true).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The CONTENT relation: SAY-class events take propositional (true) content; ASK-class events take question (false) content; other events have no propositional content.
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.rudinContent e b = match e.runtime.start with | 0 => b = true | 1 => b = true | 2 => b = true | 3 => b = true | 4 => b = false | x => False
Instances For
Verb predicates: defined as the postulate RHS so the iff-axioms
hold by rfl.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
A non-LINGMAT RESP performance: a non-linguistic, non-rising
interrogative (e.g., a wordless interrogative gesture). Its
update is askPolarQuestion, which pushes an issue without
committing. Used to falsify SAY for ASK-class events.
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.respNonLingmat = { form := Dialogue.FarkasBruce.SentenceForm.interrogative, content := fun (x : Bool) => true = true, lingmat := false }
Instances For
Witness performances #
Concrete FBPerformance witnesses with named property proofs. Each
witness pins down the exact field configuration that makes a particular
cell of the matrix felicitous, and is referenced both in rudinModel's
postulate proofs and in the IsRudinModel instance discharge.
A neutral committing declarative performance.
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.committingDecl = { form := Dialogue.FarkasBruce.SentenceForm.declarative, content := fun (x : Bool) => true = true }
Instances For
A loud committing declarative performance.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A whispered committing declarative performance.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A rising-declarative performance (RESP, not committing).
Equations
- Phenomena.Quotation.Studies.Rudin2025LI.risingDecl = { form := Dialogue.FarkasBruce.SentenceForm.declarative, content := fun (x : Bool) => true = true, rising := true }
Instances For
A loud rising-declarative performance.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A whispered rising-declarative performance.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The Rudin model: a concrete SpeechVerbs instantiation over
FBPerformance Bool with fbOntology as its performance ontology.
Each meaning postulate holds by rfl since the verb predicates
are defined as the postulate RHS.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All 30 cells of @cite{rudin-2025b}'s empirical matrix are derived from the FB-grounded model + the SpeechVerbs postulates.
Classify an English VerbEntry into the Rudin verb taxonomy.
Returns none for verbs that don't fall into the matrix (e.g.,
tell requires a recipient; think is not a speech act).
Reads directly off Fragment fields — speechActVerb,
takesQuestionBase, levinClass, and surface form — so the
classification stays in sync with Fragment edits.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Per-entry classification witnesses #
These examples pin individual Fragment verbs to the Rudin taxonomy.
Renaming or reclassifying any of these verbs in the Fragment will
break exactly the relevant witness, surfacing the inconsistency.