Documentation

Linglib.Studies.Creissels2025

Creissels (2025): Transitivity, Valency, and Voice #

[Cre25]

Formalization of key results from [Cre25], a comprehensive typological reference on transitivity, valency, and voice across the world's languages. The book proposes a unified framework based on:

This study file bridges [Cre25]'s framework (formalized in Syntax/ArgumentStructure/Alternation.lean) to existing linglib infrastructure:

Bridge: Decausativization ↔ IntransitivizationType #

§8.3.1.2: "Decausativization suppresses the referent of the initial A from participant structure and converts the initial P into the S term of an intransitive construction."

This corresponds to IntransitivizationType.anticausative in linglib's existing causation typology. Creissels prefers "decausativization" because the prefix de- transparently marks removal of the causation component, while "anticausative" misleadingly suggests a parallel with "antipassive" that doesn't hold (§8.3.1.2).

Reflexive intransitivization is NOT decausativization: the causer is coidentified with the causee (bieventive), not removed. In Creissels' terms, reflexive intransitives are a different structural operation.

Bridge: Causativization ↔ CausativeConstruction #

§8.3.1.1 defines causativization as "the nucleativization of a participant (the causer) that instigates the event denoted by the initial construction or controls its realization."

The existing CausativeConstruction type adds the fine structure: morphological complexity (lexical/morphological/periphrastic) and semantic mediation (direct/indirect). Creissels' causativization is the structural frame; linglib's CausativeConstruction fills in the parameters.

§12.1.4 gives the tripartite morphological distinction:

Causativization and decausativization are inverse operations on the causality chain (§8.3.1): causativization adds a causer in A, decausativization suppresses A from participant structure.

Bridge: Applicativization ↔ Pylkkänen's ApplType #

§14.1 distinguishes three varieties:

Pylkkänen (2008)'s high/low distinction is orthogonal to Creissels' P/D/X distinction. High applicatives introduce event-level participants (benefactives); low applicatives introduce transfer participants (recipients, sources). In Creissels' terms, high applicatives tend to produce P-applicativization (the applied phrase gets P coding), while low applicatives tend to produce D-applicativization (dative coding).

P-applicativization is valency-increasing (the applied phrase becomes a new core term).

Bridge: symmetrical voices ↔ the Voice substrate #

§8.5: symmetrical voice systems are those in which verb morphology marks the selection of a participant as the privileged syntactic term (pivot) WITHOUT AFFECTING TRANSITIVITY. This is a fundamentally different type of voice system from A/P-prominent systems (§1.3.3.3).

The Voice substrate's VoiceSystemSymmetry captures this with .symmetrical vs .asymmetrical, but doesn't encode Creissels' key insight: symmetrical voices are NOT instances of passivization, causativization, etc. — they are a distinct type that doesn't fit the nucleativization/denucleativization framework at all.

Example Toba Batak voice profiles illustrate that symmetrical systems have 2+ voices with equal morphological complexity (equipollent marking).

An A/P-prominent transitive construction (e.g., English active/passive) is an asymmetrical voice system.

Equations
Instances For

    Passivization vs Decausativization (§8.3.1.2 vs §8.3.2.1) #

    §8.3.2.1: "The maintenance of the initial A in participant structure is essential to distinguish passivization from decausativization."

    Both operations denucleativize A and yield an intransitive construction, but they differ in whether A remains in participant structure:

    This distinction is now directly encoded in ParticipantFate.

    Passivization and decausativization are structurally distinct despite both yielding intransitive constructions: they differ in whether A remains in participant structure.

    Bridge: Passivization ↔ Antipassivization structural symmetry #

    §8.3.2: passivization, antipassivization, and S-denucleativization form a natural class — all three denucleativize a core term without nucleativizing any other participant, and the denucleativized participant remains in participant structure. They differ only in which core term is targeted:

    Bridge: Reflexivization/Reciprocalization ↔ existing data #

    §8.3.3: reflexivization and reciprocalization cumulate two participant roles (A and P) into a single participant (S). They differ in whether S refers to a single individual (reflexive) or a group whose members mutually fill both roles (reciprocal).

    Both are classified as valency-decreasing in Creissels' framework — the derived construction has fewer core terms. The existing WALS Ch 106 data in Typology.lean captures the cross-linguistic formal relationship between reflexive and reciprocal markers.

    Voice Marker Stacking (§8.4) #

    §8.4: voice markers can be stacked compositionally. Example from Tswana (§8.4.1, ex. 38):

    The compositional stacking means we can model multi-marker verb forms as sequential application of ValencyAlternation operations.

    @[reducible, inline]

    A stacked voice derivation: a sequence of alternations applied in order.

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

        Portative Derivation (§8.3.7) #

        §8.3.7 identifies portative derivation as a distinct voice alternation type that cannot be reduced to either causativization or applicativization, although it shares features with both:

        Example: Caddo Ci-ʔa=d(ih)-ʔaʔ 'I will go' → Ci-ni-ʔa=d(ih)-ʔaʔ 'I will take it' (§8.3.7, ex. 33).

        Alignment Profiles (§1.3.4) #

        §1.3.4.2: most languages have a clear preference for either A-alignment (S codes like A) or P-alignment (S codes like P). Some languages (e.g., Basque, Georgian) show split-S patterns.

        Equations
        Instances For

          Russian -sja polysemy (§1.2, ex. 8) #

          The Russian verbal suffix -sja / -s' is a paradigmatic example of voice marker polysemy. It marks at least four different voice alternation types:

          (8a) reflexivization: Ivan mo-et-sja 'Ivan washes (himself)' (8b) reciprocalization: Paren' i devuška celuj-ut-sja 'The boy and the girl were kissing' (8c) passivization: Lekcija čita-et-sja professor-om 'The course is delivered by the professor' (8d) antipassivization: Sobaka kusa-et-sja 'The dog bites (people)'

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

            Russian -sja is polysemous across four voice alternation types.

            Tswana -el polysemy (§8.2) #

            The Tswana voice suffix -el (traditionally called "applicative") marks both P-nucleativization (applicativization) and A-nucleativization of obliques (non-causative A/S-nucleativization). Example:

            (12) Ki-tłàà-kwál-él-á Kítsó lò-kwâːlɔ̀ 'I'll write the letter to/for Kitso' (P-nucleativization of recipient) (13b) Nàmà í-ʃáb-él-à bò-χɔ́ːbɛ̀ 'Meat gives flavor to the porridge' (A-nucleativization of instrument → A)

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

              Causativizability and Voice Alternations #

              Ch 12 discusses restrictions on causativization. [Kre12]'s hierarchy (already formalized in MorphologicalCausation.lean) describes which verb classes can be causativized: unaccusatives > middles/ingestives > unergatives > simple transitives. This hierarchy predicts that causativization of transitive verbs (§12.3.5) often requires antipassivization to create an intransitive base first.

              Causativization of transitives may require prior antipassivization to create an intransitive base (§12.3.5). This is an instance of compositional voice marker stacking.

              Equations
              Instances For

                Cross-linguistic voice distribution (Bahrt 2021) #

                §8.3.8 reports [Bah21]'s survey of synthetic voice marking across 222 languages from all genera.

                Cross-linguistic prevalence of a voice alternation type: the share of languages with synthetic marking for it.

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

                    [Bah21] distribution data, cited in §8.3.8: the fraction of the 222 languages (all genera) with synthetic marking for each voice alternation type. Note (§8.3.8): Bahrt's "applicativization" conflates applicativization with non-causative A/S-nucleativization, so that row is broader than Creissels's pApplicativization.

                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For
                      theorem Creissels2025.causativization_most_common :
                      Option.map (fun (x : VoiceDistribution) => x.alternation.name) bahrt2021Distribution.head? = some "causativization"

                      Causativization is the most common voice alternation type cross-linguistically ([Bah21]).

                      Compositional denotation of voice stacking (§8.4) #

                      §8.4: voice markers stack, and a combination "operates on its valency properties exactly as it could operate on the valency of an underived verb form" — compositional stacking (§8.4.1). We model an alternation as a word over atomic coding-frame edits (the nucleativization/denucleativization atoms of §8.1.3), composition as the free-monoid product, and meaning as a denotation into partial transformations of coding frames. Two MonoidHoms — denote and valencyDelta — make stacking and the valency grading hold by construction.

                      End-roles (the S→P of causativization, P→S of passivization) are derived by normalizeFrame from the construction's transitivity, not stipulated — faithful to §8.1.3 (whose only atoms are nucleativization/denucleativization; the role changes are coding consequences, cf. footnote 5 §8.3.2.1).

                      This layer currently has a single consumer (this study), so per the project's graduation discipline it lives here; the FrameMap partial-transformation monoid (a Kleisli/Function.End analogue) and the denote/valencyDelta apparatus would hoist to Core/ and Alternation.lean once a second study consumes them.

                      A term's coding status in a construction (a state; ParticipantFate is the transition, derived below via fateOf).

                      Instances For

                        A coding-frame entry: a stable identity (tracked through a derivation) and its current status.

                        Instances For
                          def Creissels2025.Stacking.instDecidableEqCodedTerm.decEq (x✝ x✝¹ : CodedTerm) :
                          Decidable (x✝ = x✝¹)
                          Equations
                          Instances For
                            @[reducible, inline]

                            A coding frame (§1.3.3): the participants and their statuses.

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

                                Recompute a term's role from transitivity: the sole core term is S; a core S alongside other cores becomes P (a transitive clause has A and P, an intransitive clause has S).

                                Equations
                                Instances For

                                  Normalize end-roles after an edit — this derives the S↔P relabellings a relabel atom would otherwise stipulate.

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

                                          Some core term in the frame bears role r.

                                          Equations
                                          Instances For

                                            The atomic voice operations: Creissels's nucleativization / denucleativization (§8.1.3) plus cumulation (reflexivization, §8.3.3).

                                            Instances For
                                              def Creissels2025.Stacking.instDecidableEqAtom.decEq (x✝ x✝¹ : Atom) :
                                              Decidable (x✝ = x✝¹)
                                              Equations
                                              Instances For
                                                @[reducible, inline]

                                                An alternation as a word over atoms; composition = the free-monoid product (= voice-marker stacking, §8.4).

                                                Equations
                                                Instances For

                                                  Partial transformations of coding frames under Kleisli composition; f * g is "apply f then g" (the MulOpposite of Function.End-style order). Kept a def (not abbrev) so this Monoid is keyed on FrameMap and does not leak to bare _ → Option _. Not PFun: Part ≠ Option, and the demos below rely on Option's decidability.

                                                  Equations
                                                  Instances For
                                                    @[implicit_reducible]
                                                    Equations
                                                    • One or more equations did not get rendered due to their size.
                                                    Equations
                                                    • One or more equations did not get rendered due to their size.
                                                    Instances For

                                                      Denotation of a word as a partial coding-frame transformation. A MonoidHom, so map_mul gives compositional stacking (§8.4.1).

                                                      Equations
                                                      Instances For
                                                        def Creissels2025.Stacking.valencyDelta :
                                                        Word →* Multiplicative

                                                        The valency grading: net change in number of core terms. A MonoidHom into Multiplicative ℤ, so stacking adds deltas.

                                                        Equations
                                                        Instances For
                                                          @[simp]
                                                          theorem Creissels2025.Stacking.denote_of (a : Atom) :
                                                          denote (FreeMonoid.of a) = atomDenote a
                                                          @[simp]
                                                          theorem Creissels2025.Stacking.valencyDelta_of (a : Atom) :
                                                          valencyDelta (FreeMonoid.of a) = Multiplicative.ofAdd (atomDelta a)

                                                          Id of the participant currently in core role r (frame-relative).

                                                          Equations
                                                          Instances For

                                                            Named alternations as words (roles fall out of normalizeFrame) #

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

                                                              Forward bridges: the denotation reproduces the comparative-concept records #

                                                              These prove the word denotations agree with the schema-level ValencyAlternation records (Alternation.lean) where both are defined — the legitimate direction, not a redefinition of the schema.

                                                              Passivization obliquifies the initial A (kept in participant structure); P becomes S by normalization.

                                                              Decausativization suppresses the initial A (removed). Same net valency as passive, different A-fate — the headline §8.3.1.2 / §8.3.2.1 distinction.

                                                              Applicativization adds a new core P; on a transitive base this yields a double-P construction (§8.1.8) — no role relabelling required.

                                                              Reflexivization cumulates A and P into a single core term, which normalizes to S (§8.3.3).

                                                              Real composition for voice stacking (§8.4) #

                                                              theorem Creissels2025.Stacking.denote_stack (x y : Word) (f : CodingFrame) :
                                                              denote (x * y) f = (denote x f).bind (denote y)

                                                              Stacking is genuine composition: a word product denotes the Kleisli composite of its parts (§8.4.1). This gives the denotational counterpart of the record-level, semantics-free causativization_via_antipassive (§13).

                                                              Under stacking, "the A" is frame-relative: after causativization of an intransitive, the participant in role A is the new causer (id 1), not the original (id 0, now P). So a fate must be read off the current frame's role-holder, not a frozen "initial A".

                                                              §12.3.5: causativization via prior antipassivization is valency-neutral (antipassive −1 then causative +1), composed as one transformation.