Krifka: Commitment Space Semantics and Layered Assertive Clauses #
@cite{krifka-2015} @cite{krifka-2020}
Combines two strands of Krifka's work:
Commitment spaces: the discourse state is a tree — the root (√C) is the current CG, and continuations are proposed future states from questions. Assertion narrows every state (
C + S⊢φ); questions preserve the root and add branches (C + ?φ). Per-agent commitment slates track individual public commitments, enabling the commitment/belief separation (lying, hedging).Layered assertive clauses: four syntactic layers each contributing a distinct semantic dimension:
| Layer | Contribution | Example Modifier |
|---|---|---|
| TP | Propositional content | tense, aspect |
| JP (Judge Phrase) | Epistemic judgment | "I think", evidentials |
| ComP (Commitment Phrase) | Commitment strength | "I swear", "perhaps" |
| ActP (Act Phrase) | Speech act type | declarative, imperative |
JP and ComP are independent: "I think I swear p" vs "I swear I think p" both involve TP content p, but with different epistemic/commitment profiles.
Commitment Space Tree #
The discourse state is a CommitmentSpace tree (§4):
CommitmentSpace = { root : List Prop, continuations : List (List Prop) }
- Assert
C + S⊢φ: adds φ to root AND all continuations - Question
C + ?φ: preserves root, adds continuation (root ∩ φ), narrows existing continuations by φ - Accept: promotes first continuation to root
- Reject: prunes first continuation
The four syntactic layers of an assertive clause.
Ordered from innermost (TP, propositional content) to outermost (ActP, speech act force). Each layer contributes a distinct semantic dimension that can be independently modified.
TODO: @cite{krifka-2015} §4 also posits PolP (Polarity Phrase) between ComP and TP, hosting verum (⊢) / falsum (⊣). High negation at ComP (¬⊢φ = non-commitment to φ) vs TP negation (⊢¬φ = commitment to ¬φ) explains the bias asymmetry of negative questions: "Isn't it raining?" is biased toward rain because negation scopes over ⊢, not over φ.
- TP : ClauseLayer
Tense Phrase: propositional content
- JP : ClauseLayer
Judge Phrase: epistemic status (certainty/uncertainty)
- ComP : ClauseLayer
Commitment Phrase: commitment strength
- ActP : ClauseLayer
Act Phrase: speech act type
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Rank ordering of clause layers (innermost = 0).
Equations
Instances For
TP is the innermost layer.
Graded commitment strength, controlled by ComP modifiers.
weak: hedged ("I think p", "maybe p")standard: default declarative assertionstrong: oath formulae ("I swear p", "I promise p")
- weak : CommitmentStrength
- standard : CommitmentStrength
- strong : CommitmentStrength
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Numerical ordering of commitment strengths.
Equations
Instances For
Standard is the default.
A fully layered assertion, decomposed into the four clause layers.
Each layer is independent: the epistemic status (JP) can vary without
affecting the commitment strength (ComP), and vice versa. The actType
uses IllocutionaryMood from Core/Discourse/DiscourseRole.lean.
- content : BProp W
TP: the propositional content
- epistemicStatus : CommitmentStrength
JP: the speaker's epistemic status toward the content
- commitmentStrength : CommitmentStrength
ComP: the strength of the speaker's public commitment
- actType : Core.Discourse.IllocutionaryMood
ActP: the type of speech act performed
Instances For
Agent type for two-participant discourse.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Theories.Pragmatics.Assertion.Krifka.instBEqKAgent.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Bridge KAgent to the framework-agnostic DiscourseRole.
Both encode the same two-participant distinction.
Equations
Instances For
A commitment space (@cite{krifka-2015}, Definition 3, p.329).
A set of commitment states organized as root + continuations:
- root (√C): the current commitment state — what is actually in the common ground. All worlds compatible with the root propositions are "live".
- continuations: proposed future states, each extending the root. Added by questions, resolved by acceptance or rejection.
Krifka's key operations:
- Assert
C + S⊢φ= {s ∩ ⟦φ⟧ : s ∈ C}: narrow every state by φ - Question
C + ?φ= {√C} ∪ (C + S₂⊢φ): preserve root, propose φ - Accept: take a continuation as the new root
- Reject: prune a continuation
The tree structure captures the assertion/question asymmetry: assertions modify the root (the CG changes), while questions only add continuations (the CG is preserved until acceptance).
Root commitment state √C: propositions currently in the CG
Proposed future states. Questions add these; acceptance promotes one to root. Each continuation is a list of propositions that extends (narrows) the root.
Instances For
The empty commitment space: no commitments, no continuations.
Equations
Instances For
Assert φ: narrow every state by φ (@cite{krifka-2015}, (9), p.329).
C + S⊢φ = {s ∩ ⟦φ⟧ : s ∈ C}
Adds φ to the root AND all continuations. Any accepted continuation will also entail φ.
Equations
Instances For
Question: preserve root, propose φ (@cite{krifka-2015}, (14), p.332).
C + ?φ = {√C} ∪ (C + S₂⊢φ)
The root stays unchanged (CG preserved). A new continuation is added (root narrowed by φ), and existing continuations are also narrowed by φ. The addressee can accept (promote continuation to root) or reject (prune it).
Equations
Instances For
The space is settled: no open continuations. A settled space has no unresolved proposals.
Equations
Instances For
Accept the first continuation: it becomes the new root. The CG is updated to the accepted proposal's content.
Equations
- { root := root, continuations := c :: rest }.acceptFirst = { root := c, continuations := rest }
- x✝.acceptFirst = x✝
Instances For
Reject the first continuation: prune it. The CG is unchanged; the proposal is discarded.
Equations
- { root := root, continuations := c :: rest }.rejectFirst = { root := root, continuations := rest }
- x✝.rejectFirst = x✝
Instances For
Context set from root: worlds compatible with all root propositions.
Instances For
Accepting a simple question's sole continuation re-settles.
Root after assertion contains the asserted proposition.
Question preserves root commitments.
Krifka's discourse state: per-agent commitment slates + shared commitment space (tree).
The commitment space tracks the shared discourse structure: what's in the CG (root) and what's been proposed (continuations). Per-agent slates track individual public commitments, enabling the commitment/belief separation central to Krifka's theory.
- speakerCS : Core.Discourse.Commitment.CommitmentSlate W
Speaker's individual commitment slate
- addresseeCS : Core.Discourse.Commitment.CommitmentSlate W
Addressee's individual commitment slate
- space : CommitmentSpace W
Shared commitment space (tree): CG + proposed updates
Instances For
Initial state: no commitments, empty settled space.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Krifka's commitment operator + S₁⊢p (@cite{krifka-2015}, (9)):
speaker commits to p, narrowing the entire space.
Equations
Instances For
Accept: addressee also commits to p (after speaker's assertion).
Equations
Instances For
Assert = commit (@cite{krifka-2015}: assertion IS commitment). The space is narrowed immediately; the CG reflects the assertion.
Instances For
Question: speaker proposes φ as a continuation (@cite{krifka-2015}, (14)). The CG (root) is unchanged; a new continuation is added.
Equations
Instances For
Accept the first continuation: it becomes the new CG.
Equations
- s.acceptContinuation = { speakerCS := s.speakerCS, addresseeCS := s.addresseeCS, space := s.space.acceptFirst }
Instances For
Reject the first continuation: prune it.
Equations
- s.rejectContinuation = { speakerCS := s.speakerCS, addresseeCS := s.addresseeCS, space := s.space.rejectFirst }
Instances For
Retract: remove a commitment from the speaker's slate.
Equations
Instances For
Context set: from the commitment space root (= CG).
Equations
- s.contextSet = s.space.toContextSet
Instances For
Stability: the space is settled (no open proposals).
Instances For
ComP preserves TP content: changing commitment strength does not change the propositional content.
JP and ComP are independent: changing one does not affect the other.
Krifka's ActP layer directly uses IllocutionaryMood, grounding
the clause-type distinction in the speech act classification.
Update type for assertions.
Krifka distinguishes two fundamentally different ways an assertion can change the common ground:
informative (
·φ): eliminates worlds incompatible with φ.c + ·φ = {i | i ∈ c ∧ φ(i)}Example: "The meeting is at 5" — reduces uncertainty.performative (
•φ): changes world indices so φ becomes true.c + •φ = {i' | ∃ i ∈ c, i ⇒_φ i'}Example: "I hereby name this ship the Queen Elizabeth" — makes φ true by the act of uttering it.
This distinction is orthogonal to commitment strength (ComP) and epistemic status (JP).
- informative : UpdateType
- performative : UpdateType
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Informative update: restrict context set to worlds satisfying φ.
@cite{krifka-2020}, (7): c + ·φ = {i | i ∈ c ∧ φ(i)}
Equations
Instances For
A fully specified assertion with update type.
- updateType : UpdateType
Whether the update is informative or performative
Instances For
Default assertions are informative (the common case).
Commitment Closure (@cite{krifka-2020}, (25)): assertion immediately narrows the commitment space. The root (CG) after asserting φ is the original root with φ prepended.
In the tree model, this is definitional: assert adds φ to
all nodes including the root.
Question then accept ≈ assert (on the root): accepting a question's sole continuation yields the same CG as a direct assertion.
This connects the two modes of updating: direct assertion (speaker imposes) and question-then-accept (speaker proposes, addressee accepts). The roots match because both produce φ :: root₀.
The two discourse roles in a speech act.
Every speech act has an actor (who performs the act) and a committer (who undertakes the commitment). These can diverge:
- Assertion: actor = speaker, committer = speaker
- Monopolar question: actor = speaker, committer = addressee (the speaker proposes that the addressee commit)
This is the "one promising aspect" Krifka highlights in his conclusion: the framework naturally separates who acts from who commits, explaining why questions can be biased (the speaker chooses WHAT to propose) without being assertive (the addressee decides WHETHER to commit).
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
In assertions, speaker is both actor and committer.
Equations
Instances For
In monopolar questions, speaker acts but addressee commits.
@cite{krifka-2015}, (16): C +_{S₁} [? [⊢ p]] proposes that S₂ commit to p.
Equations
Instances For
Assertions and questions differ in who commits.
A speech act in Krifka's framework: ActP content with its discourse function (assertion vs question).
@cite{krifka-2015} clause structure: ActP > ComP > TP (three layers). The 2020 paper refines this to ActP > ComP > JP > TP.
- content : BProp W
Propositional content (TP layer)
- actType : Core.Discourse.IllocutionaryMood
Speech act type: assertion (.) or question (?)
- roles : ActorCommitter
Actor/committer assignment
Instances For
Monopolar question: proposes a single continuation where the addressee commits to φ.
C +_{S₁} [? [⊢ φ]] = {√C} ∪ (C + S₂⊢φ)
The root is preserved ({√C}) alongside the proposed continuation (C + S₂⊢φ). The addressee can accept (take the continuation) or reject (stay at root). This is what makes it a QUESTION rather than an assertion — the commitment is proposed, not imposed.
Monopolar questions are inherently biased toward the proposed continuation.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Complex speech act: conjunction or disjunction of atomic acts.
@cite{krifka-2015}, §5: question tags involve composition of speech acts. The difference between matching and reverse tags is conjunction vs disjunction:
- conj: both acts performed sequentially (matching tag)
- disj: addressee chooses one continuation (reverse tag)
- atom
{W : Type u_1}
: SpeechAct W → ComplexSpeechAct W
A single speech act.
- conj
{W : Type u_1}
: SpeechAct W → SpeechAct W → ComplexSpeechAct W
Conjunction: perform both acts sequentially. "It's raining, isn't it?" = ASSERT(rain) ∧ ASK(rain).
- disj
{W : Type u_1}
: SpeechAct W → SpeechAct W → ComplexSpeechAct W
Disjunction: offer alternative continuations. "It's raining, or isn't it?" = ASSERT(rain) ∨ ASK(¬rain).
Instances For
Extract the component speech acts from a complex speech act.
Equations
Instances For
A matching question tag: conjunction of assertion + monopolar question with same content.
"It's raining, isn't it?" = speaker asserts rain AND asks addressee to confirm. The speaker has already committed, so the question is biased toward the asserted content.
Equations
Instances For
A reverse question tag: disjunction of assertion + monopolar question with opposite content.
"It's raining, or isn't it?" = speaker offers two continuations. The addressee picks one.
Equations
- One or more equations did not get rendered due to their size.
Instances For
In a matching tag, the speaker is actor in both acts.
In a matching tag, the committers differ: speaker for assertion, addressee for question.
Matching tags are conjunctions, reverse tags are disjunctions.
A commitment space projects to a context set via its root.
A Krifka state projects to a context set via the commitment space root.
KrifkaState context set agrees with CommitmentSpace context set.