The algebraic structure of morphosyntax #
[SM25] models Distributed Morphology inside the Merge algebra of [MCB25]: morphological objects are built by the same free non-associative commutative magma as syntactic objects, differing only in labeling — an internal vertex carries the union of its children's feature bundles, so a bundle is derived from the tree rather than stipulated. On this carrier fusion is the magma operation itself, fission restricts the tree along a partition of its bundle, and impoverishment is a fission component: the paper reduces the four DM operations to fusion and fission.
Main definitions #
bundle— the feature bundle of a morphological object, the set of features at its leavesfuse,restrict— fusion as magma grafting, and the vertex-wise restriction producing the two fission outputsFeatureCorrespondence,Morphosyntactic,Matched— the syntax-morphology feature correspondence and morphosyntactic trees in labeled form: each syntactic leaf carries its datum and its inserted morphological objecttoSyntactic,msFeatures— the forgetful projection to the syntactic tree, and the total feature content of an assembly
Main statements #
bundle_fuse— the fused bundle is the union of the input bundles, an instance of the labeling law: fusion needs nothing beyond the magmabundle_restrict— a fission output's bundle is the input bundle restricted to the kept featuresrestrict_copies_shared— features shared out to both sides of the partition survive in both outputsbundle_fuse_restrict— fission followed by fusion restores the bundle, andfuse_restrict_ne— but not the treemsFeatures_fuse,msFeatures_fission_partition— at the application site, fusion and disjoint fission move the syntax-morphology boundary (toSyntacticcollapses or grows a cherry) while the total feature content is invariantmatched_fused_iff— the fused site is matched exactly when the union bundle matches the projecting datumPostSyntactic.toFinset_msFeatures_eq,Derivation.toFinset_msFeatures_subset— along the post-syntactic semigroup the total feature set is invariant; the full Distributed Morphology semigroup can only shrink it
TODO #
The workspace Hopf algebra with obliteration via the coproduct, and the colored correspondence between the syntactic and morphosyntactic algebras over the Merge operad.
References #
Morphological objects and the derived bundle labeling #
A morphological object is a nonplanar tree with single features at the
leaves and bare structural vertices elsewhere — the alphabet F ⊕ Unit,
exactly the carrier shape of syntactic objects with features in place of
the lexical items. Definition 2.1 of [SM25] labels
each internal vertex with the union of its children's bundles, so every
vertex label is determined by the leaves below it; bundle is the
root's.
Features occur only at leaves: internal vertices are structural. Restriction can leave childless or non-branching structural vertices, so arity is not constrained (the extended morphological objects of [SM25] Definition 2.8).
- leaf {F : Type u_1} (x : F ⊕ Unit) : IsMorphological (RoseTree.node x [])
- node {F : Type u_1} {cs : List (RoseTree (F ⊕ Unit))} (hne : cs ≠ []) (h : ∀ c ∈ cs, IsMorphological c) : IsMorphological (RoseTree.node (Sum.inr ()) cs)
Instances For
The multiset of leaf features of a planar tree.
Equations
- SenturiaMarcolli2025.leafFeatures t = Multiset.filterMap Sum.getLeft? t.leaves
Instances For
The feature bundle of a morphological object: the set of features at its leaves, the root label of [SM25] Definition 2.1.
Equations
- SenturiaMarcolli2025.bundle S = (Multiset.filterMap Sum.getLeft? S.leaves).toFinset
Instances For
The multiset of leaf features of a morphological object; bundle is
its underlying set. Fission with an overlapping partition duplicates
features, which only the multiset records.
Equations
- SenturiaMarcolli2025.features S = Multiset.filterMap Sum.getLeft? S.leaves
Instances For
Fusion #
Fusion merges the bundles at two adjacent leaves of the syntactic tree into one ([SM25] §5.1, Swahili negation si- fusing NEG with first-singular agreement). On morphological objects it is the magma operation itself — grafting under a fresh structural root — which is why fusion is the one DM operation available inside syntax.
Fusion of two morphological objects: the magma product, grafting both under a structural root ([SM25] Definition 5.2).
Equations
- SenturiaMarcolli2025.fuse S₁ S₂ = RoseTree.Nonplanar.node (Sum.inr ()) {S₁, S₂}
Instances For
The fused bundle is the union of the input bundles — the labeling law itself, so fusion needs nothing beyond the magma.
Fusion is additive on the feature multiset: nothing is lost, duplicated, or created.
Fission #
Fission splits one bundle into two along a partition B ∖ A = B₁ ⊔ B₂,
with the residue A copied into both outputs ([SM25]
§5.2; Ṣanʕānī Arabic discontinuous agreement, where person and number of
a single head surface as prefix and suffix). Each output is the input
tree restricted vertex-wise to the kept features: leaves outside the
kept set are deleted, structural vertices stay, and vanished subtrees
may leave non-branching structural vertices behind — the extended
morphological objects.
Keep a leaf feature iff it lies in C; structural vertices always
survive.
Equations
- SenturiaMarcolli2025.keep C (Sum.inl f) = if f ∈ C then some (Sum.inl f) else none
- SenturiaMarcolli2025.keep C (Sum.inr val) = some (Sum.inr ())
Instances For
Restriction of a morphological object to the features in C: the
construction of a single fission output ([SM25]
Definition 5.6, with C = Bᵢ ∪ A). none only for a single leaf
outside C.
Equations
Instances For
The bundle of a fission output is the input bundle restricted to the
kept features: the vertex law B_w ∩ (Bᵢ ∪ A) of
[SM25] Definition 5.6, at the root.
Multiset form of bundle_restrict: restriction filters the feature
multiset, none counting as empty.
Impoverishment as fission plus fusion #
[SM25] Propositions 5.20–5.21: impoverishment and
obliteration are not independent operations. Discarding a subbundle is
keeping one fission output (bundle_restrict with the kept part as
C); the trace-maintaining variant fissions and refuses at the same
vertex, restoring the bundle without restoring the tree. Obliteration
replaces a whole morphological object by the unit of the workspace
algebra, which lives in the second slice.
Fission followed by fusion restores the bundle: the trace-maintaining impoverishment composite leaves the feature content intact.
The tree is not restored #
Fission at a partition whose residue is copied into both outputs, then
fusion, restores the bundle but not the tree: the copied feature now
occupies a leaf on each side, so the composite has strictly more leaves.
The witness is [SM25] Example 5.7: [φ, α, β, γ] with
kept sets [φ, γ] and [φ, α, β].
Equations
- SenturiaMarcolli2025.instDecidableEqFeat x✝ y✝ = if h : SenturiaMarcolli2025.Feat.ctorIdx✝ x✝ = SenturiaMarcolli2025.Feat.ctorIdx✝ y✝ then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- SenturiaMarcolli2025.instReprFeat = { reprPrec := SenturiaMarcolli2025.instReprFeat.repr }
Restriction to {φ, γ} and {φ, α, β} followed by fusion yields a
five-leaf object: φ is realized on both sides, so the composite is not
the original four-leaf tree even though its bundle is
(bundle_fuse_restrict).
The syntax-morphology correspondence and morphosyntactic trees #
A morphosyntactic tree inserts morphological objects at the leaves of a syntactic tree, constrained by a matching rule between the bundle at the root of the inserted object and the syntactic datum at the leaf ([SM25] Definitions 2.4 and 3.4). We keep the insertion as a leaf label — syntactic datum paired with the inserted object — which carries the same data as splicing the object below the leaf.
The syntax-morphology feature correspondence: which feature bundles can be matched with which lexical items and syntactic features ([SM25] Definition 2.4). Matching is multivalued in both directions, but every syntactic datum carries some bundle.
- matching : Finset F → Λ → Prop
matching B lexiff the bundleBcan decorate a leaf carrying the syntactic datumlex. - matching_surjective (lex : Λ) : ∃ (B : Finset F), self.matching B lex
Every syntactic datum carries some bundle.
Instances For
A morphosyntactic tree in labeled form: each syntactic leaf carries
its datum together with the inserted morphological object, none where
morphology was obliterated ([SM25] Definition 3.4; the
empty insertion is Remark 5.12).
Equations
- SenturiaMarcolli2025.Morphosyntactic F Λ = RoseTree.Nonplanar (Λ × Option (RoseTree.Nonplanar (F ⊕ Unit)) ⊕ Unit)
Instances For
A leaf of a morphosyntactic tree: a syntactic datum with an optional morphological insertion.
Equations
- SenturiaMarcolli2025.insertion lex mo = RoseTree.Nonplanar.leaf (Sum.inl (lex, mo))
Instances For
The forgetful projection to the syntactic tree: drop the inserted morphology. This is [SM25] Definition 3.9's morphism of algebras over the Merge operad, in labeled form.
Equations
- SenturiaMarcolli2025.toSyntactic = RoseTree.Nonplanar.map (Sum.map Prod.fst id)
Instances For
The feature content of one leaf label.
Equations
- SenturiaMarcolli2025.insertionFeatures (Sum.inl (fst, some S)) = SenturiaMarcolli2025.features S
- SenturiaMarcolli2025.insertionFeatures (Sum.inl (fst, none)) = 0
- SenturiaMarcolli2025.insertionFeatures (Sum.inr val) = 0
Instances For
The total feature content of a morphosyntactic tree: everything its inserted morphological objects carry, with multiplicity.
Equations
- SenturiaMarcolli2025.msFeatures T = (Multiset.map SenturiaMarcolli2025.insertionFeatures (RoseTree.Nonplanar.leaves T)).sum
Instances For
Every inserted morphological object matches its leaf's syntactic datum.
Equations
- SenturiaMarcolli2025.Matched Γ T = ∀ (lex : Λ) (S : RoseTree.Nonplanar (F ⊕ Unit)), Sum.inl (lex, some S) ∈ RoseTree.Nonplanar.leaves T → Γ.matching (SenturiaMarcolli2025.bundle S) lex
Instances For
The DM operations at their application site #
The four DM operations transform the assembly of a morphosyntactic tree
at one site ([SM25] §5–§6). Fusion turns a syntactic
cherry over two insertions into a single insertion of the magma product;
fission turns one insertion into a cherry over its two restrictions.
Under toSyntactic the first collapses a cherry and the second grows
one — the movable syntax-morphology boundary — while msFeatures
records what happens to the feature content.
The application site of fusion: a syntactic cherry with insertions at both leaves.
Equations
- SenturiaMarcolli2025.fusionSite lex₁ lex₂ S₁ S₂ = RoseTree.Nonplanar.node (Sum.inr ()) {SenturiaMarcolli2025.insertion lex₁ (some S₁), SenturiaMarcolli2025.insertion lex₂ (some S₂)}
Instances For
The application site of fission on S along the kept sets C₁, C₂:
a syntactic cherry whose leaves carry the two restrictions; an empty
restriction is an empty insertion.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Fusion preserves the total feature content at the site: the fused insertion, under whichever of the two data projects (the head function's choice in [SM25] Definition 5.2), carries exactly what the cherry carried.
Fusion raises the syntax-morphology boundary: the syntactic
projection collapses from a cherry to a single leaf while the feature
content stays constant (msFeatures_fuse).
The necessary condition for fusion: the fused site is matched exactly
when the union of the two bundles matches the projecting datum
([SM25]'s condition (B_{v₁} ∪ B_{v₂}, α_v) ∈ Γ_SM
following Definition 5.2).
Fission along a disjoint cover of the bundle preserves the feature
multiset: the boundary moves down (toSyntactic grows a cherry) with no
change in feature content. An overlapping partition instead duplicates
the shared residue (restrict_copies_shared), which is how the copied
features of discontinuous agreement come to be pronounced twice.
Impoverishment at a leaf only removes features: keeping one fission output ([SM25] Proposition 5.20, first case) bounds the feature content by the original.
Obliteration removes the entire insertion while the syntactic leaf and its datum stay in place — [SM25]'s reading of Proposition 5.13, on which no morphology is inserted and the syntactic tree is untouched.
One-step transformations and the post-syntactic semigroup #
The DM operations act anywhere in a morphosyntactic tree: a step applies
a site transformation at the root of some subtree. Compositions of such
steps form [SM25] Definition 6.1's semigroups — the
post-syntactic semigroup generated by fusion and fission, and the full
Distributed Morphology semigroup adding impoverishment and obliteration.
Their action is the movable syntax-morphology boundary: PostSyntactic
preserves the total feature set exactly, and Derivation can only
shrink it.
Close a site relation under congruence: apply it at the root or inside one subtree.
- here {F : Type u_1} {Λ : Type u_2} {R : Morphosyntactic F Λ → Morphosyntactic F Λ → Prop} {T T' : Morphosyntactic F Λ} : R T T' → Step R T T'
- congr {F : Type u_1} {Λ : Type u_2} {R : Morphosyntactic F Λ → Morphosyntactic F Λ → Prop} {x : Λ × Option (RoseTree.Nonplanar (F ⊕ Unit)) ⊕ Unit} {cs : Multiset (Morphosyntactic F Λ)} {T T' : Morphosyntactic F Λ} : Step R T T' → Step R (RoseTree.Nonplanar.node x (T ::ₘ cs)) (RoseTree.Nonplanar.node x (T' ::ₘ cs))
Instances For
The fusion site relation: a syntactic cherry over two insertions rewrites to a single insertion of the magma product, projecting one of the two data.
- mk {F : Type u_1} {Λ : Type u_2} (lexHead lex₁ lex₂ : Λ) (S₁ S₂ : RoseTree.Nonplanar (F ⊕ Unit)) (hhead : lexHead = lex₁ ∨ lexHead = lex₂) : FuseAt (fusionSite lex₁ lex₂ S₁ S₂) (insertion lexHead (some (fuse S₁ S₂)))
Instances For
The fission site relation: an insertion rewrites to a syntactic
cherry over the two restrictions determined by a partition
B ∖ A = B₁ ⊔ B₂ of its bundle with copied residue A
([SM25] Definition 5.6).
- mk {F : Type u_1} [DecidableEq F] {Λ : Type u_2} (lex lex₁ lex₂ : Λ) (A B₁ B₂ : Finset F) {t : RoseTree (F ⊕ Unit)} (ht : IsMorphological t) (hlex : lex = lex₁ ∨ lex = lex₂) (h₁ : Disjoint A B₁) (h₂ : Disjoint A B₂) (h₁₂ : Disjoint B₁ B₂) (hcover : ∀ f ∈ features (RoseTree.Nonplanar.mk t), f ∈ A ∪ B₁ ∪ B₂) : FissAt (insertion lex (some (RoseTree.Nonplanar.mk t))) (fissionSite lex₁ lex₂ (B₁ ∪ A) (B₂ ∪ A) (RoseTree.Nonplanar.mk t))
Instances For
The impoverishment site relation: the insertion is replaced by one of its fission outputs.
- mk {F : Type u_1} [DecidableEq F] {Λ : Type u_2} (lex : Λ) (kept : Finset F) {t : RoseTree (F ⊕ Unit)} (ht : IsMorphological t) : ImpovAt (insertion lex (some (RoseTree.Nonplanar.mk t))) (insertion lex (restrict kept (RoseTree.Nonplanar.mk t)))
Instances For
The obliteration site relation: the insertion is emptied, the syntactic leaf and its datum staying in place.
- mk {F : Type u_1} {Λ : Type u_2} (lex : Λ) (mo : Option (RoseTree.Nonplanar (F ⊕ Unit))) : OblitAt (insertion lex mo) (insertion lex none)
Instances For
Fission duplicates exactly the copied residue: with B ∖ A
partitioned into B₁ ⊔ B₂ and A copied to both sides, the site's
feature multiset grows by precisely the A-part. A = ∅ recovers
content invariance (msFeatures_fission_partition).
Feature bookkeeping along the semigroups #
A site-level feature-content identity transfers through the congruence closure.
Fusion moves the boundary up: one syntactic leaf fewer, wherever it applies.
Fission moves the boundary down: one syntactic leaf more, wherever it applies.
The two semigroups and the movable boundary #
The post-syntactic semigroup: compositions of fusion and fission steps ([SM25] Definition 6.1).
Equations
- One or more equations did not get rendered due to their size.
Instances For
A derivation of the full Distributed Morphology semigroup: post-syntactic steps together with impoverishment and obliteration ([SM25] Definition 6.1).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The movable boundary, feature side: the post-syntactic semigroup never creates or destroys a feature. Any composition of fusions and fissions leaves the total feature set invariant — the operations only move the boundary between syntax and morphology.
Along the full Distributed Morphology semigroup the feature set can only shrink: fusion and fission preserve it, impoverishment and obliteration delete.
The economy of bivalent features #
[SM25] Remark 2.3: with n feature categories and
three valuations (+, −, unvalued), the bivalent inventory has n + 3
generating objects against the 3 n features a privative encoding
needs, strictly fewer for more than one category.