Hyman 2006: Word-prosodic typology #
@cite{hyman-2006}
Hyman, L. M. (2006). Word-prosodic typology. Phonology 23, 225--257.
Core claims formalized #
Two prototypes, not three: The highest-level typological cut distinguishes two prototypical word-prosodic systems — tone (T) and stress accent (SA) — not three. Pitch accent (PA) is not a coherent third type but a "pick and choose" combination of properties from both prototypes.
Definition of tone (def. 3): "A language with tone is one in which an indication of pitch enters into the lexical realisation of at least some morphemes." Tone is featural and paradigmatic.
Definition of stress accent (def. 5): "A language with stress accent is one in which there is an indication of word-level metrical structure" meeting two criteria: (a) OBLIGATORINESS — every lexical word has at least one primary stress; (b) CULMINATIVITY — every lexical word has at most one primary stress. OBLHEAD is the more important criterion.
2×2 typology (Table I): Since T and SA are independent properties, languages may have both (+T+SA), either (+T−SA, −T+SA), or neither (−T−SA).
Non-definitional properties: Additional properties cluster with but do not define the SA prototype: privativity, subordination, demarcation, rhythmicity.
Table II: Restricted H-tone systems independently attest all four combinations of obligatoriness × culminativity, showing these are orthogonal.
OBLHEAD as the deepest cut: The most significant typological cut is between systems satisfying OBLHEAD and those that do not.
Connection to @cite{lionnet-2025} #
Hyman's tone prototype encompasses all systems where pitch enters lexical realization. @cite{lionnet-2025} shows this category splits further into tone-based (paradigmatic H/L: Yoruba, Mandarin) and register-based (syntagmatic downstep: Drubea, Numèè). This enrichment is orthogonal to Hyman's SA dimension.
Hyman's two independent binary word-prosodic properties.
A language's word-prosodic profile is characterized by two independent Boolean dimensions, not a single categorical type. This is the key insight: T and SA can freely co-occur (Table I).
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
The four cells of Table I.
- toneAndStress : ProsodicQuadrant
- toneOnly : ProsodicQuadrant
- stressOnly : ProsodicQuadrant
- neither : ProsodicQuadrant
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
The two central definitional criteria for stress accent (def. 5).
OBLIGATORINESS is more important: it is "an absolute universal — definitional — of a SA system" (p. 232). Culminativity may be violated in some alleged PA systems.
- obligatoriness : Bool
(5a) Every lexical word has at least one syllable with primary stress.
- culminativity : Bool
(5b) Every lexical word has at most one syllable with primary stress.
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
A system satisfying both criteria is a prototypical SA system.
Equations
- c.isPrototypicalSA = (c.obligatoriness && c.culminativity)
Instances For
OBLHEAD requires both obligatoriness and syllable-targeting.
Equations
Instances For
Additional properties that cluster with the SA and T prototypes but are not definitional.
SA-clustering properties include privativity, subordination, demarcation, and rhythmicity. T-clustering properties include being paradigmatic and identifying TBU tones.
- privativity : Bool
Privativity: stress is present vs. absent, not graded (p. 234).
- subordination : Bool
Subordination: one stress may be subordinated to another (p. 234).
- demarcation : Bool
Demarcation: stress fixed on initial/final/etc. signals word boundary (p. 234).
- rhythmicity : Bool
Rhythmicity: echo-stresses occur on every other syllable (p. 234).
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.
- Phenomena.Tone.Studies.Hyman2006.instBEqClusteringProperties.beq x✝¹ x✝ = false
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Prototypical SA systems exhibit all four clustering properties.
Equations
Instances For
Language entry for the 2×2 typology (Table I, p. 237).
- name : String
- profile : WordProsodicProfile
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Table I entries: representative languages for each quadrant.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All four quadrants are attested in Table I.
Table II (p. 245): restricted H-tone systems attest all four combinations of obligatoriness and culminativity.
- name : String
- criteria : StressAccentCriteria
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Table II entries.
Equations
- One or more equations did not get rendered due to their size.
Instances For
All four combinations of obligatoriness × culminativity are attested.
Only Kinga satisfies both definitional criteria for SA.
"PA-like" properties (13): three distinct things that have been called pitch accent. These do not cohere into a single prototype.
- abstractDifferent : PALikeProperty
(13a) Underlying prosody abstractly different from surface realisations.
- combinesToneAndStress : PALikeProperty
(13b) System combines tone and stress.
- restrictedTone : PALikeProperty
(13c) Restricted, sparse, or privative tone (e.g., /H/ vs. ∅).
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Hyman's key argument: languages called "PA" freely pick and choose between these properties. No single definition of PA can be given that parallels the definitions for T (3) and SA (5).
The claim is that PA reduces to T (since pitch enters lexical realization), with additional SA-like properties possibly co-occurring. Hyman reclassifies Tokyo Japanese, Somali, and Western Basque as [+T, −SA] — i.e., tonal.
Equations
Instances For
Instances For
Equations
Instances For
All three classic "PA" languages are reclassified as +T, −SA.
WALS F13A maps onto Hyman's binary tone dimension:
noTones → −T; simpleToneSystem/complexToneSystem → +T.
Equations
Instances For
WALS F14A maps onto Hyman's binary stress dimension:
noFixed (free stress) → +SA; fixed locations → +SA;
absence from WALS survey → indeterminate.
Note: WALS F14A only classifies fixed stress location.
Free stress languages (English, Russian) appear as noFixed,
which still indicates +SA. The −SA case (Bella Coola, French)
is not directly captured by WALS F14A.
Equations
Instances For
Derive a partial word-prosodic profile from WALS data.
The stress dimension is only available when the language has a WALS F14A entry. Languages absent from F14A may or may not have stress — their profiles require independent evidence.
Equations
- One or more equations did not get rendered due to their size.
Instances For
English: −tone (WALS none), +stress (WALS noFixed = free stress).
Finnish: −tone, +stress (WALS initial = fixed stress).
Yoruba: +tone (WALS complex), −stress (absent from WALS F14A).
Japanese: +tone (WALS simple), −stress (absent from WALS F14A).
Mandarin: +tone (WALS complex), +stress (WALS noFixed).
Zulu: +tone (WALS simple), +stress (WALS penultimate).
Three of the four quadrants are attested among the 16 PhonProfile
languages. The −T−SA quadrant is NOT attested because WALS F14A
encodes stress location but not stress absence: languages
like French appear as noFixed (= free stress = +SA), even though
Hyman classifies French as −SA.
This demonstrates the inherent limitation of deriving Hyman's framework from WALS data alone.
French is classified as −T−SA in Hyman's Table I, but WALS F14A
records it with noFixed stress (= free stress), which our
profileFromPhon maps to +SA. This mismatch arises because
WALS F14A does not distinguish "free word stress" (English) from
"no word stress" (French — stress is phrase-final, not word-level).
This is not a bug in the formalization but a documented limitation of the WALS→Hyman bridge. The −SA classification requires language-specific analysis beyond WALS data.
Whether a WordProsodicType counts as tonal under Hyman's
definition (3): "an indication of pitch enters into the lexical
realisation of at least some morphemes."
@cite{lionnet-2025} enriches Hyman's tone prototype by splitting it into tone-based (paradigmatic H/L) and register-based (syntagmatic h/l). Both sub-types satisfy definition (3).
Equations
- Phenomena.Tone.Studies.Hyman2006.isTonalUnderHyman Theories.Phonology.Autosegmental.RegisterTier.WordProsodicType.toneBased = true
- Phenomena.Tone.Studies.Hyman2006.isTonalUnderHyman Theories.Phonology.Autosegmental.RegisterTier.WordProsodicType.registerBased = true
- Phenomena.Tone.Studies.Hyman2006.isTonalUnderHyman Theories.Phonology.Autosegmental.RegisterTier.WordProsodicType.mixed = true
- Phenomena.Tone.Studies.Hyman2006.isTonalUnderHyman Theories.Phonology.Autosegmental.RegisterTier.WordProsodicType.stressAccent = false
Instances For
Both tone-based and register-based systems count as +tone under Hyman's definition (3).
Stress accent systems are −tone under Hyman's definition.
All four WordProsodicType values from @cite{lionnet-2025} map
consistently to Hyman's binary +/−tone.
The WALS 3-way tone classification (F13A) does not capture the tone-based vs. register-based distinction. Both Yoruba (tone-based) and Drubea (register-based) would be classified as "simple" or "complex" in WALS, losing the syntagmatic vs. paradigmatic distinction that @cite{lionnet-2025} identifies.
Drubea's word-prosodic profile under Hyman's framework: +T, −SA.
+T: register features (downstep) enter lexical realization — the minimal pairs in @cite{lionnet-2025} ex. 4 demonstrate this directly. −SA: no obligatory metrical head; the prosodic system is purely register-based with no stress accent.
This places Drubea in the same quadrant as Yoruba, Igbo, and the reclassified "PA" languages (Tokyo Japanese, Somali).
Instances For
Drubea is in the +T−SA quadrant (tone only).
Drubea's profile is consistent with Yoruba's (both +T, −SA from different sub-types of tone).
Drubea lacks stress accent — it is not an OBLHEAD system.
Hyman's stress culminativity (5b) and Drubea's register culminativity (@cite{lionnet-2025}) share the name but apply to different domains:
- Stress culminativity (Hyman): at most one primary stress per word. Definitional for SA systems. Violated in some alleged PA systems (Creek, Seneca — Table II).
- Register culminativity (Lionnet): at most one
lfeature per native stem. A distributional constraint on register features, not a metrical property.
Both are "at most one X per domain" constraints, but the X differs (metrical prominence vs. register feature) and the domain differs (word vs. stem).
- stressPerWord : CulminativityDomain
- registerPerStem : CulminativityDomain
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
The two culminativity constraints are formally parallel but apply to different phonological objects.
Hyman's final proposal (§7): the most significant typological cut is OBLHEAD — whether every word obligatorily has a metrical head.
OBLHEAD = OBLIGATORINESS: every lexical word has at least one syllable marked for the highest degree of metrical prominence. This is the unique defining property of SA systems.
Systems satisfying OBLHEAD are SA (possibly with co-occurring T). Systems not satisfying OBLHEAD include tone-only systems and −T−SA systems (e.g., Bella Coola, which lacks syllables and therefore cannot have OBLHEAD).
Instances For
OBLHEAD separates SA (±T) from non-SA systems.
OBLHEAD partitions Table I: languages with +SA are OBLHEAD systems, regardless of whether they also have tone.
Every Table I language with −SA also has −OBLHEAD.