Vocabularies over containment hierarchies: Elsewhere insertion #
The realizational engine behind [Bob12]'s comparative-suppletion
generalizations, stated over an arbitrary n-grade containment
hierarchy. An ExponenceRule realizes the initial span [0, spans] of
the hierarchy (the root, possibly fused with the first spans heads —
the portmanteau exponence of [Bob12]'s ch. 5, via Fusion (196)
or Radkevich's Vocabulary Insertion Principle (197)), optionally
conditioned on the presence of a higher head. Selection at each grade
is by the Elsewhere Condition ([Kip73]): the most specific
applicable rule wins, where Pāṇinian specificity is derived as
applicability-set inclusion
(ExponenceRule.moreSpecific_iff_threshold_le) — over a linear
hierarchy applicability sets are nested upward sets, so the Elsewhere
ordering is total.
Each of [Bob12]'s generalizations is a theorem with its own hypothesis set, mirroring the book's cost accounting:
- CSG1 / *ABA (
isContiguous_realize): Antihomophony alone — the book's ch. 2 derivation from containment + Elsewhere + the antihomophony assumption (44). The homophony loophole is real: a non-antihomophonous vocabulary generating surface ABA is exhibited inStudies/Bobaljik2012.lean. - Completeness (
isContiguous_iff_generable): generable = contiguous over terminal antihomophonous vocabularies. - The plateau (
realize_const_of_terminal_adjacent): terminal rules + adjacent contexts generate only{AAA, ABB}— the Bobaljik-minus-portmanteaux fragment, which over-excludes the attested ABC (Latin bon-mel-opt-); portmanteau rules repair it. - CSG2 / *AAB (
csg2): Antihomophony +Grounded(the book's markedness condition (202)). The book blocks the two AAB routes by two mechanisms — adjacency for root-rules conditioned across an intervening head (190a), condition (202) for context-sensitive portmanteaux (201); under the threshold encoding both reduce to downward closure of the threshold set, so one hypothesis suffices. - ABC requires a portmanteau (
exists_portmanteau_of_ne; [Bob12] §5.3.1, the degree-domain consequence generalized there as (199)): under adjacency, superlative-grade root allomorphy distinct from the comparative grade arises only via a portmanteau rule.
Main declarations #
ExponenceRule n F— rule of exponence: exponent, exponed span[0, spans], optional conditioning headTerminal,Adjacent,Antihomophonous,Grounded— the axiom Props on vocabularies, à la cartewinner,realize— Elsewhere selection and the realized patternisContiguous_realize,isContiguous_iff_generable,realize_const_of_terminal_adjacent,csg2,exists_portmanteau_of_ne
The dual Superset engine lives in
Morphology/Containment/Superset.lean; synthetic/analytic structure
(Merger) in Morphology/Containment/Merger.lean; the n = 3 degree
instantiations with the book's worked vocabularies in
Studies/Bobaljik2012.lean.
Rules of exponence and derived specificity #
A rule of exponence ([Bob12]'s term — Matthews's
exponence, used in the realizational sense of [Stu01]) over an
n-grade containment hierarchy. The rule realizes the initial span
[0, spans] — the root when spans = 0, a root+heads portmanteau
when spans > 0 — and applies only when the (optional) conditioning
head context is present in the structure. DM vocabulary items are
the Terminal-restricted special case; nanosyntax lexical entries
share the context-free span format but pair it with Superset-based
selection (Morphology/Containment/Superset.lean) rather than this
file's containment-directed AppliesAt, so the insertion semantics
differs.
Latin ([Bob12] (204)): bon is ⟨bon, 0, none⟩, mel- is
⟨mel, 0, some 1⟩, the portmanteau opt- is ⟨opt, 1, some 2⟩.
- exponent : F
The phonological exponent inserted for the span.
- spans : Fin n
Upper end of the exponed initial span
[0, spans]. - context : Option (Fin n)
Head whose presence conditions the rule, if any.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
The least grade at which the rule is applicable: everything the rule mentions — exponed span and conditioning context — must be contained in the structure.
Instances For
A rule applies at grade g when grade g's structure contains
everything the rule mentions.
Instances For
Equations
Pāṇinian specificity: it is at least as specific as jt when it
applies in a subset of the structures jt applies in.
Equations
- it.MoreSpecific jt = ∀ ⦃g : Fin n⦄, it.AppliesAt g → jt.AppliesAt g
Instances For
Over a linear containment hierarchy, applicability sets are nested upward sets, so derived specificity is threshold comparison — the Elsewhere ordering ([Kip73]) is total and needs no stipulated specificity ranking.
Well-formedness conditions on vocabularies #
Each generalization below hypothesizes exactly the conditions it
needs; vocabularies violating a condition witness the corresponding
unattested pattern (see the worked examples in
Studies/Bobaljik2012.lean).
No portmanteaux: every rule expones the bare root.
Equations
- Morphology.Containment.Terminal v = ∀ it ∈ v, ↑it.spans = 0
Instances For
Equations
- Morphology.Containment.instDecidableTerminal v = id inferInstance
Conditioning heads are adjacent to the exponed span — [Bob12]'s (tentatively adopted) adjacency condition on contextual allomorphy.
Equations
- Morphology.Containment.Adjacent v = ∀ it ∈ v, ∀ (c : Fin n), it.context = some c → ↑c = ↑it.spans + 1
Instances For
Equations
- Morphology.Containment.instDecidableAdjacent v = id inferInstance
Distinct rules carry distinct exponents — [Bob12]'s Antihomophony assumption (44), closing the loophole of a surface-ABA pattern that is really an ABC with accidental A ≡ C homophony. Stated as pairwise antihomophony, a mild strengthening of the book's default-vs-contextual formulation; all worked vocabularies satisfy it.
Equations
- Morphology.Containment.Antihomophonous v = ∀ it ∈ v, ∀ jt ∈ v, it.exponent = jt.exponent → it = jt
Instances For
Equations
- Morphology.Containment.instDecidableAntihomophonousOfDecidableEq v = id inferInstance
[Bob12]'s markedness condition (202): a context-sensitive
rule of exponence involving a node requires a context-free rule
involving that node. A rule is context-free for the node [0, k]
when nothing it mentions extends beyond k — i.e. its threshold is
k (the book's fn. 15 rewrites Latin mel- / __]CMPR] as the
node-level context-free GOOD, CMPR → mel-CMPR) — so (202) says the
vocabulary's threshold set is downward closed. On Adjacent
vocabularies this is exactly the book's condition; on non-adjacent
ones it is a mild strengthening (it also covers the skipped-head items
the book independently excludes by adjacency).
Equations
- Morphology.Containment.Grounded v = ∀ it ∈ v, ∀ k < it.threshold, ∃ jt ∈ v, jt.threshold = k
Instances For
Equations
- Morphology.Containment.instDecidableGrounded v = id inferInstance
Elsewhere selection #
The rules applicable at grade g, in vocabulary order.
Equations
- Morphology.Containment.applicable v g = List.filter (fun (it : Morphology.Containment.ExponenceRule n F) => decide (it.threshold ≤ g)) v
Instances For
The greatest applicable threshold at grade g — the specificity
level of the Elsewhere winner, ⊥ when nothing applies.
Equations
- Morphology.Containment.maxThreshold v g = (List.map Morphology.Containment.ExponenceRule.threshold (Morphology.Containment.applicable v g)).maximum
Instances For
The key transfer lemma: a winning threshold persists downward as long as it stays applicable. With monotone applicability this is what makes Elsewhere selection plateau between grades.
The Elsewhere winner at grade g: the first rule attaining the
greatest applicable threshold. By
ExponenceRule.moreSpecific_iff_threshold_le, this is the most specific
applicable rule.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The realized pattern: at each grade, the Elsewhere winner's
exponent (none when no rule applies — a paradigm gap).
Equations
Instances For
CSG1: realization is contiguous #
[Bob12] ch. 2: with antihomophonous rules, the Elsewhere
competition over a containment hierarchy cannot generate ABA — with
only two distinct listed root forms, no ordering of the rules yields
an ABA pattern (p. 35). Formally: maxThreshold is the
monotone score, the winner is a function of it, and antihomophony
makes exponents injective in the winner — so realization factors as
monotone-then-injective and Basic.lean's composition principle
applies (here inlined as the sandwich argument).
The plateau: terminal adjacency generates only {AAA, ABB} #
Capping all thresholds makes realization constant above the cap. With
terminal rules and adjacent contexts the cap is the first head — this
is the "Bobaljik-minus-portmanteaux" engine, which forces the
comparative and superlative cells to coincide and so over-excludes
the attested ABC patterns (Latin bon- mel- opt-). Portmanteau items
(0 < spans) are what license ABC; see exists_portmanteau_of_ne.
Terminal rules with adjacent contexts have thresholds at most the first head, so the comparative and superlative cells coincide: only AAA and ABB root patterns are generable.
Completeness: generable = contiguous #
The earliest grade sharing g's form.
Equations
- Morphology.Containment.firstOcc p g = {j : Fin n | p j = p g}.min' ⋯
Instances For
The canonical vocabulary of a pattern: one rule per form, introduced at the form's first grade and conditioned on it.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A pattern is Elsewhere-generable: some terminal antihomophonous vocabulary realizes it in full.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Generable = contiguous. A fully realized pattern arises from Elsewhere insertion over a terminal antihomophonous vocabulary iff it is contiguous: the forward direction is the canonical-vocabulary construction, the backward direction CSG1.
Three-grade hierarchies: *AAB and the portmanteau prediction #
*CSG2 / AAB exclusion ([Bob12] ch. 5). Over the
three-grade degree hierarchy, if the positive and comparative cells
agree and the superlative cell is realized, all three agree:
good – gooder – *best is not generable. Hypotheses: antihomophony
plus Grounded (the book's markedness condition (202)); the
threshold-set downward closure makes both of the book's AAB routes —
the skipped-head root rule (190a) and the context-sensitive
portmanteau (201) — fail for the same reason, a threshold gap at the
comparative grade.
ABC requires a portmanteau ([Bob12] §5.3.1, the
degree-domain consequence generalized there as (199)): under
adjacency, root allomorphy at the superlative grade distinct from the
comparative grade arises only via a portmanteau — the winning rule
must expone more than the bare root (Latin opt-, Welsh gor-).