Harbour (2016): Impossible Persons — the partition calculus #
Person categories are derived, not primitive: they are the cells of a
partition of the π-lattice induced by two features, ±author and
±participant. Crucially — and this is the thesis of [Har16a] (Ch. 4.2.3,
Ch. 9) — the features denote operations on person lattices, not predicates:
their values are actions ⊕ (disjoint addition) and ⊖ (joint subtraction),
"not truth functors, which would reduce the features to first-order predicates,
but actions by and on the person lattices."
The operators themselves (⊕/⊖/lexComp/nePowerset) are reusable substrate in
Syntax.Minimalist.Phi.Lattice; this file builds the partition construction on
top and checks the empirical results, which come out by construction rather than
being stipulated:
- the ontology is the speaker
i, heareru, and otherso; referents are sets of these (Finset Ω), and the three feature lattices are powerset sublatticesℒau ⊆ ℒpt ⊆ ℒπ(the egocentric chain, [Har16a] p. 74); - a feature acts on a collection:
+F(G) = G ⊕ ℒF(pairwise join, (17)) and−F(G) = G ⊖ ℒF(stripℒF's maximum, (19)); - cells are made disjoint by Lexical Complementarity ((31): a cell properly
containing a sibling is confined to the difference) and the ∅-winnowing of
the variable head
φ(§4.4); and - composition is noncommutative (Ch. 4.2.3), so the order of the two features
yields the standard tripartition (participant outermost: the two
−participantcells coincide) or the quadripartition (author outermost: inclusive/exclusive split).
Main declarations #
refine/cellsOf— Lexical Complementarity over a sibling family + ∅-winnowing, the partition construction overPhi.Lattice's⊕/⊖/actoperators.Examples— the concrete three-element ontology{i, u, o} = Fin 3, where the partition cells live and the derived theorems aredecide-checked.signOf— Harbour's±author/±participantsigns for a CysouwCategory(theory-laden; cf. the neutralPerson.Category.toFeatures).
Main results (all derived, none stipulated) #
inclusive_ne_exclusive,exclusive_excludes_addressee,inclusive_has_both— inclusive (+author +participant) and exclusive (+author −participant) come out disjoint, with the exclusive genuinely−participant. This is exactly the distinction a Boolean-membership encoding (the foil [Har16a] argues against) collapses.quad_covers/quad_disjoint— the cells coverℒπand are pairwise disjoint.tri_card/quad_card— composition order gives 3 vs 4 cells (noncommutativity).five_partitions— the two-feature inventory's five parameter settings yield five distinct partitions ([Har16a] Table 4.1).inclusive_has_no_singleton— the inclusive holds no singleton, so the singular realizes only three persons (no four-way singular).
The number side ([Har14a]; [Har16a] Ch. 6) is the Phi.Recursion/Corbett2000
bridge.
The partition construction (over Phi.Lattice's operators) #
Lexical Complementarity over a sibling family + the φ domain restriction
([Har16a] (31), §4.4): confine a cell c by removing every sibling cell
properly contained in it (Phi.Lattice.lexComp across the family), then winnow the
empty set (there are no empty persons).
Equations
- Harbour2016.refine siblings c = (List.foldl (fun (acc d : Finset (Finset Ω)) => if d ⊂ c then acc \ d else acc) c siblings).erase ∅
Instances For
The cells of a partition, given the raw (pre-refinement) denotations of each sign assignment: refine each against the whole family, collected as a set.
Equations
- Harbour2016.cellsOf raws = (List.map (Harbour2016.refine raws) raws).toFinset
Instances For
The three-element ontology {i, u, o} and the derived results #
Harbour's abbreviation i_o/iu_o/u_o/o_o collapses the π-lattice into four
shape-classes; the minimal ontology realizing all four is {i, u, o}, taken here
as Fin 3 with i = 0, u = 1, o = 2. A single o suffices for the person
partition (the four cells are fixed by has i?/has u?, not the others' count); the
3rd-person group {o, o'} is not modelled, and no result below depends on it. All
theorems are kernel-checked by decide on this finite model.
The author lattice ℒau = {{i}} (powerset of {i}, minus ∅).
Equations
- Harbour2016.Examples.ℒau = {{0}}
Instances For
The participant lattice ℒpt = {{i}, {u}, {i,u}} (powerset of {i,u}, minus ∅).
Equations
- Harbour2016.Examples.ℒpt = {{0}, {1}, {0, 1}}
Instances For
The π-lattice: all nonempty referent-sets.
Equations
- Harbour2016.Examples.ℒπ = Minimalist.Phi.Lattice.nePowerset Finset.univ
Instances For
Raw cell of the quadripartition for sign sa of author, sp of participant
(author outermost, participant innermost).
Equations
Instances For
The four raw quadripartition cells.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A refined quadripartition cell.
Equations
Instances For
1st-person inclusive = +author +participant.
Equations
Instances For
1st-person exclusive = +author −participant.
Equations
Instances For
2nd person = −author +participant.
Equations
Instances For
3rd person = −author −participant.
Equations
- Harbour2016.Examples.thirdP = Harbour2016.Examples.quadCell false false
Instances For
Inclusive and exclusive come out as distinct, disjoint cells by construction:
The exclusive never contains the addressee u — it is genuinely −participant
in Harbour's sense, the case a Boolean-membership decomposition collapses.
The exclusive always contains the speaker i.
The inclusive always contains both speaker and addressee.
Inclusive ≠ exclusive — they do not collapse to one feature bundle.
The quadripartition has four cells.
The calculus certifies the canonical inventory: over the
referent lattice, the four quadripartition cells are exactly the
regions Person.interp assigns to the four quadripartition values
(speaker 0, addressee 1). The generative system and the
analytical inventory (Features/Person/Basic.lean) agree cell by
cell.
Raw cell of the tripartition for sign sp of participant, sa of author
(participant outermost). The two −participant cells coincide.
Equations
Instances For
The (four) raw tripartition cells.
Equations
- Harbour2016.Examples.triRaws = [Harbour2016.Examples.triRaw true true, Harbour2016.Examples.triRaw true false, Harbour2016.Examples.triRaw false true, Harbour2016.Examples.triRaw false false]
Instances For
Composition order matters: with participant outermost, the two −participant
specifications coincide, so the tripartition has only three cells — whereas the
quadripartition has four. This is the noncommutativity of feature composition
([Har16a] Ch. 4.2.3).
Monopartition: no features active.
Instances For
Author bipartition: {±author}.
Equations
- One or more equations did not get rendered due to their size.
Participant bipartition: {±participant}.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The standard tripartition as a set of cells.
Instances For
The quadripartition as a set of cells.
Instances For
The five parameter settings of [Har16a] Table 4.1 — ∅, {±author},
{±participant}, and the two composition orders of {±author, ±participant} — yield five
distinct partitions (monopartition, the two bipartitions, the tripartition, the
quadripartition). This theorem checks the distinctness; that these five settings exhaust
the two-feature parameter space is the enumeration of Table 4.1 itself.
The singleton referents (atoms): {i}, {u}, {o}.
Equations
- Harbour2016.Examples.singletons = {{0}, {1}, {2}}
Instances For
No four-way singular person ([Har16a] Ch. 4): the inclusive — the only cell distinguishing the quadripartition from the tripartition — contains no singleton referent, so the singular realizes only three persons (the inclusive/exclusive split collapses).
Every inclusive referent set has ≥ 2 members (it contains both speaker and addressee) —
the elementwise form of inclusive_has_no_singleton.
The phi kernel: a shared form, not a shared denotation ([Har16a] Ch. 9) #
Harbour's "phi kernel" is the cluster of structural properties that person and number
features share: both are operations (not predicates), both are bivalent, and
both have freely-ordered composition. It is not a 1 ↔ sg / 2 ↔ du / 3 ↔ pl
identification — §9.5.1 stresses that the two families' values denote differently
(person's ± are ⊕/⊖ on the person lattices; number's + is the identity map and
− a type-flexible ¬). What they share at the level of the abstract ContainmentPair
carrier is the containment hierarchy: the person hierarchy 1 > 2 > 3 and the number
hierarchy sg > du > pl are both the specification ordering maximal > intermediate > minimal.
The person hierarchy 1 > 2 > 3 is the specification ordering of ContainmentPair
(maximal > intermediate > minimal), connecting [Har16a]'s lattice approach to the
hierarchy in [BR09]'s Cyclic Agree and [Pre14]'s relativized
probing.
The number hierarchy sg > du > pl is the same specification ordering — the
structural reflection of the phi kernel, not an identification of the categories.
The person hierarchy from ContainmentPair.specLevel agrees with the segment-count
hierarchy of [BR09]'s Cyclic Agree: in the standard geometry, specLevel + 1 = segment count (1st: 2/[π, participant, speaker]; 2nd: 1/[π, participant]; 3rd: 0/[π]).
The person hierarchy is one hierarchy in two formalizations — featural (spec level) and
configurational (probe depth).
Bridge to Corbett (2000): attested number systems are generable ([Har14a]) #
Every attested number system from [Cor00] is a subset of what some well-formed
Harbour configuration generates (a language may not lexicalize every category its geometry
affords). The number side uses the three-feature ±atomic/±minimal/±additive calculus
of [Har14a] (developed at chapter length in [Har16a] Ch. 6), faithfully distinct
from the two-feature person calculus above.
Every attested number system's values are generated by some well-formed Harbour
configuration: Pirahã (∅ → no number), English/Russian/Japanese (±atomic → sg/pl),
Upper Sorbian/Slovene (±atomic, ±minimal → sg/du/pl), Bayso (±atomic, ±additive →
sg/pl/pauc), Larike and Lihir (with ±minimal recursion).
MIN/AUG systems from [Har14a] Table 3.
The Harbour configurations used in the bridge are all well-formed.
Harbour's sign decomposition of the Cysouw categories ([Har16a] Table 4.3) #
The neutral Person.Category.toFeatures underdetermines the group categories
(excl/minIncl/augIncl all ⟨true,true⟩). Harbour's operational signs distinguish
them; that distinction is this theory's commitment, derived from the partition above. A
dedicated Sign carrier is used rather than Person.Features, because the
exclusive's +author −participant is exactly the combination the neutral type's wellFormed
invariant (SAP containment: author ⟹ participant) rejects — for operations, not SAP-membership
predicates, that invariant does not apply ([Har16a] Ch. 9).
A Harbour ±author/±participant sign — bivalent feature values ([Har16a]
Ch. 9), distinct from the SAP-membership Person.Features.
- participant : Bool
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
Equations
- Harbour2016.instReprSign = { reprPrec := Harbour2016.instReprSign.repr }
Harbour's signs for a Cysouw Category ([Har16a] Table 4.3). The 1st-person
exclusive — and the singular speaker .s1, which Harbour's quadripartition lumps into
the exclusive cell i_o (p. 96) — is +author −participant, so it does not collapse
with the inclusive +author +participant, unlike the neutral Category.toFeatures.
Equations
- Harbour2016.signOf Person.Category.minIncl = { author := true, participant := true }
- Harbour2016.signOf Person.Category.augIncl = { author := true, participant := true }
- Harbour2016.signOf Person.Category.s1 = { author := true, participant := false }
- Harbour2016.signOf Person.Category.excl = { author := true, participant := false }
- Harbour2016.signOf Person.Category.s2 = { author := false, participant := true }
- Harbour2016.signOf Person.Category.secondGrp = { author := false, participant := true }
- Harbour2016.signOf Person.Category.s3 = { author := false, participant := false }
- Harbour2016.signOf Person.Category.thirdGrp = { author := false, participant := false }
Instances For
Harbour's signs distinguish exclusive from inclusive, where the neutral membership
decomposition (Category.toFeatures) collapses them (cf. Examples.inclusive_ne_exclusive).
Application: the Tamil clusivity contrast through the Pronoun API #
A lexical pronoun entry feeds Harbour's signs by composing Pronoun.category — the
[Cys09] category a person/number/clusivity triple realizes — with signOf. Tamil's
clusivity-marked 1pl forms naam (inclusive) and naangaL (exclusive) land on distinct signs,
where the neutral Category.toFeatures collapses both 1pl categories to ⟨true, true⟩: the
distinction [Har16a]'s decomposition exists to draw, here discharged on real Fragment
entries rather than a stipulated example.
The Harbour sign a pronoun realizes: its [Cys09] Category (via the Pronoun API's
Pronoun.category) decomposed by signOf. none when the φ-features underdetermine a
category. The bridge from a lexical pronoun entry to [Har16a]'s ±author/±participant
operations.
Equations
- Harbour2016.harbourSign p = Option.map Harbour2016.signOf p.category
Instances For
naam (1pl inclusive) realizes the inclusive sign +author +participant.
naangaL (1pl exclusive) realizes the exclusive sign +author −participant.
Harbour's signs distinguish naam from naangaL through the Pronoun API.
The neutral membership decomposition collapses them: both 1pl categories map to the same
Category.toFeatures, so only Harbour's operational signs keep naam and naangaL apart.
Person and number feature carriers are the same containment-pair
skeleton, composed through the ContainmentPair hub: 1st ↔ singular,
2nd ↔ dual, 3rd ↔ plural. One edge of the φ-feature iso-web; equates
carriers, not denotations ([Har16a] §9.5.1 — the two domains'
feature values denote differently).