Command Relations on Abstract Trees #
[BP90]'s algebraic theory of command relations.
A command relation generated by a property P is the set of node pairs
(a, b) such that every P-node properly dominating a also dominates b.
Different syntactic theories pick different P (branching nodes, maximal
projections, S-nodes, dependency heads), and the Intersection Theorem
shows the map P ↦ C_P is antitone — converting ∪ of properties into
∩ of command relations.
This is the core insight that lets the same lattice structure subsume c-command (Minimalism), o-command (HPSG), d-command (Dependency Grammar), and s-command ([Lan69]).
Main Definitions #
commandRelation T P— pairs(a, b)where every P-upper-bound ofadominatesb.branchingNodes T/cCommand T— the branching-node generating property and [Rei76]'s c-command it generates.commandByRelation T R— relation-based variant generalizing the property-based definition.mateRelation T P— symmetric closure:C_P ∩ (C_P)⁻¹.maximalGenerator T R— largest relation generating the sameC_R.
Main Theorems #
intersection_theorem—C_P ∩ C_Q = C_{P∪Q}.command_antitone—P ⊆ Q → C_Q ⊆ C_P.configurational_equivalence— when upper bounds coincide, command relations agree (explains why theories converge on configurational clauses).command_ambidextrous— every node either has a P-upper-bound or commands everything.command_bounded— adding the root to P does not changeC_P.command_fair— the fairness/transitivity-failure condition.relation_union_theorem/_reverse— full B&P Theorem 9.
References #
[BP90] "A theory of command relations".
Command relation generated by property P ([BP90] Definition 3).
C_P = {(a, b) | ∀ x ∈ UB(a, P). x dominates b}.
a P-commands b iff every P-node properly dominating a also dominates b.
Equations
- Core.Order.commandRelation T P = {ab : Node × Node | ∀ x ∈ Core.Order.upperBounds T ab.1 P, x ≤ ab.2}
Instances For
Theorem 1 (Intersection Theorem) of [BP90]:
C_P ∩ C_Q = C_{P ∪ Q}.
Union of properties gives intersection of command relations.
The map P ↦ C_P is antitone (order-reversing).
Maximal property: all nodes.
Equations
Instances For
Empty property.
Equations
Instances For
IDc-command: C_{all nodes} — the most restrictive command relation.
Equations
Instances For
Universal command: C_∅ — the least restrictive (everything commands everything).
Instances For
Mother relation ([BP90] Definition 9): a is the
mother of b iff both are tree nodes and a properly dominates b
with no tree node strictly between — the minimal proper dominator.
Membership in T.nodes is required throughout because upperBounds
and the ambient order range over the whole carrier.
Equations
Instances For
Branching nodes ([BP90] P₅, eq. (19), via the
Definition 9 mother relation): nodes that are the mother of two
distinct nodes. On Branching.toTreeOrder positions this coincides
with the carrier-geometric Branching.isBranchingAt (≥ 2 children).
Equations
- Core.Order.branchingNodes T = {n : Node | ∃ (a : Node) (b : Node), a ≠ b ∧ Core.Order.mother T n a ∧ Core.Order.mother T n b}
Instances For
IDc-command is the bottom of the lattice: IDc ⊆ C_P for all P ⊆ T.nodes.
Universal command is the top: C_P ⊆ Universal for all P.
All command relations are reflexive on tree nodes.
All command relations satisfy descent on the second argument.
Configurational Equivalence Corollary:
If the upper bounds of node a are the same for properties P and Q,
then C_P and C_Q agree on all pairs starting from a.
This formalizes the configurational assumption: different theories (using different P's) agree when their upper bounds coincide.
If P and Q have the same upper bounds at a, C_P(a, –) = C_Q(a, –).
Configurational Clause Condition:
For subject position s, if there is exactly one node x properly
dominating s, and x ∈ P ↔ x ∈ Q, then C_P and C_Q agree on
pairs from s.
In a standard clause [S [NP_subj] [VP …]], the only node properly
dominating the subject NP is S. So any P containing S agrees with any
Q containing S.
Lattice Structure #
The set of command relations on a tree.
Equations
- Core.Order.CommandRels T = {C : Set (Node × Node) | ∃ (P : Set Node), C = Core.Order.commandRelation T P}
Instances For
The Intersection Theorem restated: P ↦ C_P converts ⊔ to ⊓.
Generalized intersection theorem for arbitrary unions.
The command map as an order-reversing function Set Node →o (Set (Node × Node))ᵒᵈ.
Equations
- Core.Order.commandMap T = { toFun := fun (P : Set Node) => OrderDual.toDual (Core.Order.commandRelation T P), monotone' := ⋯ }
Instances For
The command map is order-reversing.
A command relation C_P is ambidextrous iff for all a:
either ∃x. x ∈ UB(a, P) or (a, b) ∈ C_P for all b.
[BP90] Theorem 3: All command relations are ambidextrous.
Boundedness ([BP90] Theorem 4): Adding a root node to the
generating property does not alter the command relation: C_P = C_{P ∪ {r}}.
If UB(a, P) is nonempty and (a, b) ∈ C_P, then ∃x ∈ UB(a, P). x dominates b.
Fairness witness: ¬(a C_P c) implies ∃ x ∈ UB(a, P). ¬(x dom c).
Contrapositive of the command definition.
Fairness ([BP90] Theorem 6):
(a C b) ∧ (b C c) ∧ ¬(a C c) → ∀d. (a C d) → b dominates d.
Mate relation: M_P = C_P ∩ (C_P)⁻¹.
Two nodes are P-mates iff they mutually P-command each other. Examples: clause-mates (S-command), co-arguments (NP-command).
Equations
- Core.Order.mateRelation T P = {ab : Node × Node | (ab.1, ab.2) ∈ Core.Order.commandRelation T P ∧ (ab.2, ab.1) ∈ Core.Order.commandRelation T P}
Instances For
Descent ([BP90] Theorem 5) restated:
C_P is closed under descent on the second argument.
Constituency ([BP90] Theorem 7, p. 30): every
command domain C_P(a, ·) (restricted to tree nodes) is the cone of a
single node — Reinhart's "first branching node" rendered
order-theoretically. The witness is the greatest element of
UB(a, P), which exists when UB is nonempty (the CAC makes UB
a chain, so "greatest" is well-defined); when UB(a, P) is empty,
C_P(a, ·) is universal on the tree and the witness is the root.
Stated as: ∃ x, ∀ b ∈ T.nodes, (a, b) ∈ commandRelation T P ↔ x ≤ b.
The hypothesis hmax asks for the greatest element when UB is
nonempty — a one-line consequence of finiteness, which B&P assume but
which we state explicitly here so the theorem holds for the abstract
TreeOrder (which need not be finite).
The "first branching node" reading: when P = branchingNodes T, the
greatest element of UB(a, P) is the lowest branching ancestor of a,
and Reinhart's c-command domain is exactly its cone.
Embeddability (simple form): if x properly dominates a, then commands
transfer through that upper bound.
Embeddability with CAC ([BP90] Theorem 8):
If every P-upper-bound of b either properly dominates a or sits between
a and c on the dominance path, then (a, c) ∈ C_P implies (b, c) ∈ C_P.
A command relation generated by a binary relation rather than a property.
C_R(a, b) iff ∀x. (a R x) → (x dominates b).
Equations
- Core.Order.commandByRelation T R = {ab : Node × Node | ∀ (x : Node), R ab.1 x → x ≤ ab.2}
Instances For
The property-based command is a special case of relation-based.
Command equivalence ([BP90] Definition 20):
R ~ S iff C_R = C_S.
Equations
- Core.Order.commandEquivalent T R S = (Core.Order.commandByRelation T R = Core.Order.commandByRelation T S)
Instances For
Maximal generator ([BP90] Definition 21):
R̂ = ⋃ {S | S ⊇ R ∧ S ~ R}.
The largest relation generating the same command relation as R.
Equations
- Core.Order.maximalGenerator T R a x = ∃ (S : Node → Node → Prop), (∀ (a' x' : Node), R a' x' → S a' x') ∧ Core.Order.commandEquivalent T R S ∧ S a x
Instances For
Maximal generator contains the original relation.
Key lemma for the Union Theorem: non-minimal upper bounds are in the maximal generator.
Maximal generator is command-equivalent to original.
Intersection Theorem for Relations: C_R ∩ C_S = C_{R ∪ S}.
Union Theorem ([BP90] Theorem 9, forward):
C_R ∪ C_S ⊆ C_{R̂ ∩ Ŝ}.
Union Theorem ([BP90] Theorem 9, reverse) — uses CAC.
The image of the command map: all command relations on T.
Equations
- Core.Order.commandImage T = Set.range (Core.Order.commandRelation T)
Instances For
Command relations are closed under arbitrary nonempty intersection.
The command relations form a closure system.