Argument Structure Constructions #
@cite{goldberg-1995} @cite{goldberg-shirtz-2025}
CxG's argument structure constructions and their decomposition into Müller's three universal schemata.
@cite{mueller-2013} argues "both directions right": the three universal schemata capture fully abstract constructions (ditransitive, caused-motion, resultative), but partially open and lexically specified constructions are irreducible phrasal patterns that only CxG can capture.
Key claims #
- Fully abstract constructions decompose into sequences of Head-Complement and Head-Specifier steps
- Partially open constructions (PAL, let alone, WXDY) are irreducible — they cannot be decomposed into the three schemata
- This is CxG's unique contribution: phrasal constructions beyond the schemata
Construction slots and argument frames #
A slot in an argument structure construction.
Each slot specifies a syntactic category and a semantic role for one participant in the construction's event structure.
- cat : UD.UPOS
Syntactic category of this slot (NP, V, PP, etc.)
- role : String
Semantic role label
- isHead : Bool
Whether this slot is the head of the construction
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
An argument structure construction with explicit slot structure.
This extends the basic Construction with a decomposed argument frame,
enabling formal analysis of how the construction relates to the three
universal combination schemata.
The semanticContribution field captures which meaning components
(@cite{levin-1993}) the construction adds independently of the verb
(@cite{goldberg-1995}). When a verb fuses with a construction, the
composed meaning = verb.meaningComponents.fuse cxn.semanticContribution.
This is how constructions can license alternation behavior that verbs
lack in isolation — e.g., the resultative adds CoS + causation, enabling
the causative alternation for manner verbs (@cite{levin-2026}).
- construction : Construction
The underlying construction
- slots : List ConstructionSlot
The argument frame: ordered list of slots
At least one slot should be the head
- semanticContribution : MeaningComponents
What meaning components this construction contributes independently of the verb. Defaults to
.none(no augmentation).
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Concrete argument structure constructions #
Ditransitive construction: [Subj V Obj1 Obj2]. "X CAUSES Y to RECEIVE Z" (e.g., "She gave him a book").
Equations
- One or more equations did not get rendered due to their size.
Instances For
Caused-motion construction: [Subj V Obj Obl]. "X CAUSES Y to MOVE to Z" (e.g., "She sneezed the napkin off the table"). Contributes motion + causation: verbs that lexicalize neither (like sneeze) acquire both from the construction (@cite{goldberg-1995} p. 152–179).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Resultative construction: [Subj V Obj Pred]. "X CAUSES Y to BECOME Z" (e.g., "She hammered the metal flat"). Contributes CoS + causation: manner verbs that lexicalize neither acquire both from the construction (@cite{rappaport-hovav-levin-1998}; @cite{levin-2026} §3). This is what enables the causative alternation for verbs like push that lack it in isolation.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Intransitive motion construction: [Subj V Obl]. "X MOVES to Y" (e.g., "The ball rolled down the hill").
Equations
- One or more equations did not get rendered due to their size.
Instances For
Conative construction: [Subj V Obl_at]. "X DIRECTS ACTION at Y" (e.g., "Sam kicked at Bill"). The verb designates the intended result of the directed action; the at-PP marks the target without entailing contact (@cite{goldberg-1995} p. 3–4, 63–64).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Decomposition into combination schemata #
Decompose a fully abstract construction into a sequence of combination steps.
For a construction with slots [Subj, V, Obj1, Obj2]:
- V + Obj2 → V' (Head-Complement)
- V' + Obj1 → V'' (Head-Complement)
- Subj + V'' → S (Head-Specifier)
The head slot determines which combinations are complements vs specifier.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A construction is fully compositional if it has specificity fullyAbstract
and no construction-specific pragmatic function.
This is a proxy for @cite{mueller-2013}'s structural criterion (whether the construction can be analyzed as a sequence of headed binary combinations). The proxy works because fully abstract constructions without pragmatic functions have no idiosyncratic form–meaning pairings that would resist decomposition into the three universal schemata.
Equations
Instances For
Core theorems #
Ditransitive decomposes into Head-Specifier + Head-Complement + Head-Complement.
The ditransitive [Subj V Obj1 Obj2] decomposes as:
- V + Obj2 → V' (Head-Complement)
- V' + Obj1 → V'' (Head-Complement)
- Subj + V'' → S (Head-Specifier)
Caused-motion decomposes into Head-Specifier + Head-Complement + Head-Complement.
Resultative decomposes into Head-Specifier + Head-Complement + Head-Complement.
Fully abstract constructions without pragmatic functions are fully compositional.
PAL construction is NOT fully compositional.
PAL is a phrasal construction where a phrase fills a word-level slot. This form-function pairing cannot be captured by the three schemata alone — it requires construction-specific knowledge.
Let alone construction is NOT fully compositional.
Let alone is a formal idiom with paired focus, scalar entailment, and NPI licensing requirements. These semantic/pragmatic properties cannot be derived from Head-Complement + Head-Specifier + Head-Filler.
Müller's "both directions right" (§3): the three schemata handle fully abstract constructions, but CxG's phrasal constructions are irreducible.
This formalizes the biconditional:
- Fully abstract → fully compositional (covered by universal schemata)
- There exist constructions that are not fully compositional (requires CxG)
Conative decomposes into Head-Specifier + Head-Complement.
Polysemy families (@cite{goldberg-1995} §3.3.2, I_P links) #
A polysemy family groups constructions that share one syntactic frame
but differ in meaning. The shared form is enforced by construction —
all senses are generated from the same slots, making it impossible
for a polysemy extension to silently diverge in syntax.
A polysemy family: one argument frame, multiple meanings.
All constructions in a family share the same slots definitionally —
there is no way to create an extension with different syntax. The
polysemy links (I_P) are derived, not manually assembled.
- name : String
Name of the construction family
- form : String
Human-readable form description
- slots : List ConstructionSlot
The shared argument frame
At least one slot is the head
- centralMeaning : String
Central sense meaning
Extended senses: (extension name, meaning, overridden properties)
Instances For
The central sense as an ArgStructureConstruction.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Build an extension construction. Uses the family's slots — shared
by construction, not by assertion.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All extension constructions.
Equations
Instances For
All constructions (central + extensions).
Equations
Instances For
Derive polysemy links from the family structure.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Central construction uses the family's slots (definitionally true).
Every extension uses the family's slots (definitionally true). This is the structural enforcement: shared syntax is impossible to violate because it follows from the definition, not from a proof.
All members decompose identically (same slots → same decomposition).
Ditransitive polysemy network (@cite{goldberg-1995} pp. 75–77) #
The ditransitive is not a single construction but a family of six related senses connected by polysemy links (I_P). Each sense inherits the ditransitive's syntactic form [Subj V Obj Obj₂] but differs in the semantic relation between the event participants.
The ditransitive polysemy family: six senses sharing one argument frame (@cite{goldberg-1995} pp. 75–77).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The family's slots match the standalone ditransitive definition.
Derived polysemy links.
Equations
Instances For
Subpart link (I_S) from caused-motion to intransitive motion (@cite{goldberg-1995} p. 78): the intransitive motion construction is a proper subpart of the caused-motion construction.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Metaphorical extension link (I_M) from caused-motion to resultative (@cite{goldberg-1995} pp. 81–84): the resultative is a metaphorical extension of caused-motion via the systematic metaphor motion → change, location → state.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All polysemy links have link type I_P.
Subpart link has link type I_S.
Metaphorical link has link type I_M.
All four link types are instantiated across the network.
Verb–construction fusion #
@cite{goldberg-1995}'s central claim: argument structure constructions are independent form–meaning pairings. When a verb appears in a construction, its meaning fuses with the construction's meaning. The composed meaning can have properties neither has alone.
At the level of @cite{levin-1993} meaning components, fusion is componentwise OR: if either the verb or the construction contributes a component, the composed meaning has it. This simple mechanism derives construction-dependent alternation behavior (@cite{levin-2026}):
- push alone:
{−CoS, +contact, +motion, −causation}→ no causative alternation - push + resultative:
{+CoS, +contact, +motion, +causation}→ causative alternation predicted
The construction adds what the verb lacks; predictedAlternation on the
fused result gives the correct prediction without any new alternation logic.
The composed meaning of a verb in an argument structure construction. Verb root semantics fused with the construction's semantic contribution.
Equations
- ConstructionGrammar.composedMeaning verbMC cxn = verbMC.fuse cxn.semanticContribution
Instances For
Whether an alternation is predicted for a verb in a construction.
Generalizes MeaningComponents.predictedAlternation to construction contexts.
Equations
- ConstructionGrammar.predictedAlternationInConstruction verbMC cxn alt = (ConstructionGrammar.composedMeaning verbMC cxn).predictedAlternation alt
Instances For
Core theorems: constructions that don't augment #
Ditransitive contributes nothing beyond the verb (.none).
With no augmentation, the composed meaning equals the verb's own.
Core theorems: constructions that augment #
The resultative adds CoS + causation.
The caused-motion construction adds motion + causation.
Key derivation: construction-dependent alternation #
The payoff: predictedAlternation on fused components derives that manner
verbs participate in the causative alternation inside the resultative, even
though they cannot outside it. No new alternation logic is needed — the
existing component-based prediction (mc.changeOfState && mc.causation)
fires on the fused result.
A pure manner verb (no CoS, no causation) cannot alternate alone.
A pure manner verb in the resultative CAN alternate: the construction adds the CoS and causation the verb lacks.
Concrete instance: hit-class components + resultative → causative alternation.
Concrete instance: hit-class components alone → no causative alternation.
Multiple alternation flips from a single construction #
The key architectural insight: fusing a construction's components with a verb's components can flip multiple alternation predictions simultaneously. The resultative adds CoS + causation, which unlocks not just causativeInchoative but also middle, instrumentSubject, and the resultative alternation itself — all from the same mechanism, with no new alternation logic.
This is the formal payoff of Goldbergian fusion (@cite{goldberg-1995}): constructions don't just license one new alternation — they systematically augment the verb's meaning component profile, and every alternation whose required components are now satisfied becomes available.
Hit-class verbs alone: no middle, no instrumentSubject, no resultative alternation.
Hit-class in resultative: ALL FOUR component-derived alternations flip. The resultative adds CoS + causation → fused = ⟨true, true, true, true, false, false⟩. This unlocks causativeInchoative, middle, instrumentSubject, AND resultative.
Conative stays true: hit already has contact + motion, and fusing preserves them.
Caused-motion fusion #
The caused-motion construction adds motion + causation. For touch-class verbs (pure contact, no motion), this unlocks the conative alternation (requires contact + motion) and the instrument subject alternation (requires causation).
Touch alone: ⟨false, true, false, false, false, false⟩ — only BPPA (contact)
Touch + caused-motion: ⟨false, true, true, true, false, false⟩ — conative + instrumentSubject too
Touch verbs alone: no conative, no instrumentSubject.
Touch + caused-motion: conative AND instrumentSubject flip to true. Motion + causation from the construction fill exactly what touch lacks.
Touch + caused-motion: BPPA stays true (contact preserved by fusion).
Manner-of-motion verbs in the resultative #
Manner-of-motion verbs (⟨false, false, true, false, false, true⟩) have motion
but no CoS or causation. In the resultative, they acquire both — unlocking
causativeInchoative, middle, instrumentSubject, and resultative.
Manner-of-motion verbs alone: no CI, no middle, no instrumentSubject.
Manner-of-motion + resultative: CI, middle, and instrumentSubject all flip.
Constructional augmentation summary #
Each construction unlocks a characteristic set of alternations by augmenting the verb's meaning components. The table below summarizes what each construction contributes and which alternations it enables for verbs that lack the relevant components:
| Construction | Adds | Unlocks |
|---|---|---|
| Resultative | CoS + causation | CI, middle, instrumentSubject, resultative |
| Caused-motion | motion + causation | conative (if +contact), instrumentSubject |
| Ditransitive | (nothing) | (nothing) |
These predictions are all derived from the same predictedAlternation function —
no construction-specific alternation logic exists. The construction simply changes
the input to the general prediction function.
Ditransitive adds nothing: hit verbs stay blocked in all alternations that are blocked in isolation.
Instrument specification survives fusion: cut-class verbs remain blocked from causativeInchoative and resultative even inside the resultative construction, because instrumentSpec = true persists through componentwise OR.