Hayes (1989): Compensatory Lengthening in Moraic Phonology #
@cite{hayes-1989}
Bruce Hayes. "Compensatory Lengthening in Moraic Phonology." Linguistic Inquiry 20(2): 253–306.
This study file formalizes the empirical core of Hayes 1989: the typology of compensatory lengthening (CL) and its three central arguments for moraic theory over segmental prosodic theories (X theory, CV theory).
Core Claims Formalized #
CL is governed by a prosodic frame provided by moraic structure (not by constraints on association line rearrangements).
Onset Deletion Asymmetry: CL from coda deletion is common (38 rules, 26 languages); CL from onset deletion is unattested. Moraic theory derives this from the universal non-moraicity of onsets.
Weight Prerequisite: CL occurs only in languages with a syllable weight distinction. Moraic theory derives this from language-specific moraic structure: only languages with bimoraic syllables have morae to strand.
Moraic Conservation: CL conserves total mora count. This follows automatically from the representations — no stipulation needed.
Languages Covered #
- Latin (s-deletion: *kasnus → ka:nus; onset s-deletion: *smereo → mereo)
- Middle English (vowel loss: ⟨talə⟩ → [ta:l])
- Lardil (no WBP: CL impossible — derived from
syllableToMoraic) - Estonian (trimoraic syllables, Q1/Q2/Q3 quantity system)
Latin underlying form *kasnus 'gray': σ₁ = ⟨kas⟩ (C=onset, a=nucleus[1μ], s=coda[1μ with WBP]) σ₂ = ⟨nus⟩
With WBP, the coda ⟨s⟩ bears one mora, making σ₁ heavy.
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
- Hayes1989.kasnus = { syllables := [Hayes1989.kasnus_σ₁, Hayes1989.kasnus_σ₂] }
Instances For
σ₁ of *kasnus is heavy (2 morae: nucleus + coda with WBP).
After ⟨s⟩-deletion from σ₁, one mora is stranded.
After spreading, the vowel ⟨a⟩ becomes long (2 morae). Result: σ₁ = [kaː] with 2 morae — still heavy.
Equations
- Hayes1989.kaanus_σ₁ = match Phonology.Moraic.CL.deleteMoraic Hayes1989.kasnus_σ₁ 1 with | (σ_del, stranded) => Phonology.Moraic.CL.spreadToFill σ_del stranded Phonology.Moraic.CL.SpreadDir.left
Instances For
Latin *kosmis → ko:mis 'courteous': same pattern.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Latin *smereo → mereo: ⟨s⟩ deletes word-initially (onset position). Since onset ⟨s⟩ has no mora, no CL occurs.
Equations
- Hayes1989.smereo_σ₁ = { onset := [Hayes1989.s✝], moraic := [{ seg := Hayes1989.e✝, morae := Phonology.Moraic.MoraCount.one }] }
Instances For
Onset deletion does not change the mora count.
The mora count after onset deletion is still 1 (light syllable).
Middle English ⟨talə⟩ 'tale' (original disyllabic form): σ₁ = ⟨ta⟩ (open, light), σ₂ = ⟨lə⟩ (open, light).
When word-final schwa deletes, Parasitic Delinking removes σ₂'s structure, stranding a mora. Spreading from the left fills it, lengthening ⟨a⟩.
Equations
- Hayes1989.tale_σ₁ = { onset := [Hayes1989.t✝], moraic := [{ seg := Hayes1989.a✝, morae := Phonology.Moraic.MoraCount.one }] }
Instances For
Equations
- Hayes1989.tale_σ₂ = { onset := [Hayes1989.l✝], moraic := [{ seg := Hayes1989.e✝, morae := Phonology.Moraic.MoraCount.one }] }
Instances For
Equations
- Hayes1989.tale_input = { syllables := [Hayes1989.tale_σ₁, Hayes1989.tale_σ₂] }
Instances For
Input ⟨talə⟩ has 2 total morae (one per syllable).
After schwa deletion from σ₂, one mora is stranded.
CL result: ⟨a⟩ becomes long, ⟨l⟩ resyllabifies as coda. Output σ = [taːl] with 2 morae.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Conservation: input total morae = output morae.
In a language without WBP (e.g., Lardil), syllableToMoraic produces
non-moraic codas. Deleting the coda strands zero morae → no CL.
Equations
- Hayes1989.lardil_cvc = Phonology.Moraic.syllableToMoraic { wbp := false } { onset := [Hayes1989.t✝], nucleus := [Hayes1989.a✝], coda := [Hayes1989.k✝] }
Instances For
In a language with WBP (e.g., Latin), syllableToMoraic produces
moraic codas. Deleting the coda strands one mora → CL is possible.
Equations
- Hayes1989.latin_cvc = Phonology.Moraic.syllableToMoraic { wbp := true } { onset := [Hayes1989.t✝], nucleus := [Hayes1989.a✝], coda := [Hayes1989.k✝] }
Instances For
The weight prerequisite: the difference between Latin (CL possible) and Lardil (CL impossible) is exactly the WBP parameter.
Estonian Q1/Q2/Q3 (short/long/overlong) syllables demonstrate the three-way weight distinction that moraic theory encodes directly as 1μ/2μ/3μ.
Equations
- Hayes1989.estonian_q1 = { onset := [Hayes1989.k✝], moraic := [{ seg := Hayes1989.a✝, morae := Phonology.Moraic.MoraCount.one }] }
Instances For
Equations
- Hayes1989.estonian_q2 = { onset := [Hayes1989.k✝], moraic := [{ seg := Hayes1989.a✝, morae := Phonology.Moraic.MoraCount.two }] }
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Q3 → Q2 grade shift: removing the third mora. The moraic account: Q3 has 3 morae, Q2 has 2. The shift is simply "remove the third mora," which automatically eliminates gemination when the third mora belonged to a geminate consonant.
Estonian gemination loss: Q3 ⟨paːt.ti⟩ → Q2 ⟨paː.ti⟩. σ₁ goes from 3μ to 2μ; the geminate loses its second mora.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Hayes1989.paa_ti_q2 = { onset := [Hayes1989.k✝], moraic := [{ seg := Hayes1989.a✝, morae := Phonology.Moraic.MoraCount.two }] }
Instances For
The full phonological pipeline for Latin ka:nus after CL: moraic syllabification → weight profile → prosodic word.
This demonstrates the chain that moraic theory creates:
segments + WBP → MoraicSyllable → SyllWeight → PrWd
CL output: σ₁ = [ka:] (bimoraic = heavy), σ₂ = [nus] (bimoraic = heavy). Weight profile: [H, H].
Equations
- Hayes1989.kaanus_form = { syllables := [Hayes1989.kaanus_σ₁, Hayes1989.kasnus_σ₂] }
Instances For
CL output has the weight profile [heavy, heavy].
CL output satisfies the bimoraic minimal word constraint (4μ ≥ 2μ).
Middle English: CL preserves the bimoraic minimum across syllable restructuring. Input ⟨talə⟩ = [L, L] (2μ); output [ta:l] = [H] (2μ). Both satisfy the bimoraic minimum.
This follows from Moraic Conservation (Rule (64), @cite{hayes-1989}): CL does not change total mora count, so it cannot cause a minimal word violation that wasn't already present.