Beckman & Pierrehumbert (1986) @cite{beckman-pierrehumbert-1986}

Intonational Structure in Japanese and English. Phonology Yearbook 3: 255–309.

Core Contributions

This paper establishes the prosodic hierarchy above the word — accentual phrase (AP), intermediate phrase (ip), intonation phrase (IP) — via cross-linguistic comparison of Japanese and English intonation. It introduced three key analytical innovations:

1. Accentual phrase: the domain of pitch accent distribution, delimited by a phrasal H and boundary L in Japanese. At most one pitch accent per AP (culminativity at the phrasal level).

2. Catathesis: pitch range compression triggered by bitonal pitch accents. The domain is the intermediate phrase. Blocked by ip boundaries. Chaining produces descending F0 staircases.

3. Intermediate phrase: the domain of catathesis. In Japanese, 1–3 APs, bounded by pause or glottalization + L boundary tone. In English, decomposed from @cite{pierrehumbert-1980}'s single "phrase accent" into an ip-terminal tone (phrase accent) distinct from the IP boundary tone.

The paper also demonstrates that Japanese tonal patterns are much sparser than earlier autosegmental accounts assumed — closer to English in their distribution of tones to tone-bearing units.

Bridge to RegisterTier

Catathesis is formalized as register-based downstep applied to the intonation domain: each bitonal pitch accent within an intermediate phrase contributes a register l feature, producing cumulative terracing via realizePitch (@cite{snider-1999}, @cite{lionnet-2025}). The ip boundary resets the register, preventing catathesis from propagating across phrases.

structure BeckmanPierrehumbert1986.AccentualPhrase : Type

An accentual phrase: the lowest level of prosodic phrasing defined by the intonation pattern.

In Japanese, delimited by a phrasal H and boundary L tone. Contains at most one pitch accent. An unaccented AP has the rising pitch shape (L → phrasal H) but no accent HL fall (§2.2).

In English, the AP is less firmly established. It corresponds to the domain of a single pitch accent plus the surrounding material up to the next accent or phrase boundary (§2.4).

Accentedness is derived from the accent field: an AP is accented iff its accent is non-null.

• accent : Features.Prosody.PitchAccent

  Pitch accent type (null if unaccented)

• nWords : ℕ

  Number of words grouped in this phrase

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instReprAccentualPhrase : Repr AccentualPhrase

Equations

• BeckmanPierrehumbert1986.instReprAccentualPhrase = { reprPrec := BeckmanPierrehumbert1986.instReprAccentualPhrase.repr }

def BeckmanPierrehumbert1986.instReprAccentualPhrase.repr : AccentualPhrase → ℕ → Std.Format

Equations

• One or more equations did not get rendered due to their size.

def BeckmanPierrehumbert1986.instDecidableEqAccentualPhrase.decEq (x✝ x✝¹ : AccentualPhrase) : Decidable (x✝ = x✝¹)

Equations

• One or more equations did not get rendered due to their size.

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instDecidableEqAccentualPhrase : DecidableEq AccentualPhrase

Equations

• BeckmanPierrehumbert1986.instDecidableEqAccentualPhrase = BeckmanPierrehumbert1986.instDecidableEqAccentualPhrase.decEq

def BeckmanPierrehumbert1986.AccentualPhrase.isAccented (ap : AccentualPhrase) : Bool

An AP is accented iff it has a non-null accent.

Equations

• ap.isAccented = (ap.accent != Features.Prosody.PitchAccent.null)

def BeckmanPierrehumbert1986.catathesisToRegisterSpecs (aps : List AccentualPhrase) : List Phonology.Autosegmental.RegisterTier.TRN

Convert a sequence of APs into register specifications.

Each bitonal accent contributes a register l (downstep) feature; non-bitonal or null accents contribute no register feature (registerless). This bridges catathesis to @cite{snider-1999}'s RegisterTier: the descending staircase in catathesis chains (Fig. 11) is the same terracing effect produced by cumulative register l features.

Equations

• One or more equations did not get rendered due to their size.

theorem BeckmanPierrehumbert1986.catathesis_terracing (n : ℤ) : Phonology.Autosegmental.RegisterTier.realizePitch n [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.downstep] = [n - 1, n - 1 - 1]

Catathesis produces terracing (§3, Fig. 11): a sequence of bitonal accents within an ip produces a descending staircase. This follows from realizePitch applied to the register specs.

theorem BeckmanPierrehumbert1986.catathesis_lowers (n : ℤ) : (Phonology.Autosegmental.RegisterTier.realizePitch n [Phonology.Autosegmental.RegisterTier.TRN.downstep]).head? = some (n - 1)

Catathesis lowers pitch: after one bitonal accent, the pitch is strictly lower than the starting level.

def BeckmanPierrehumbert1986.catathesisCount (aps : List AccentualPhrase) : ℕ

Count catathesis applications in a sequence of APs.

Equations

• BeckmanPierrehumbert1986.catathesisCount aps = (List.filter (fun (x : BeckmanPierrehumbert1986.AccentualPhrase) => x.accent.isBitonal) aps).length

theorem BeckmanPierrehumbert1986.no_bitonal_no_catathesis (aps : List AccentualPhrase) (h : ∀ ap ∈ aps, ap.accent.isBitonal = false) : catathesisCount aps = 0

An ip with no bitonal accents has zero catathesis — no pitch compression occurs.

theorem BeckmanPierrehumbert1986.unaccented_no_catathesis : Features.Prosody.PitchAccent.null.isBitonal = false

Catathesis is triggered by the accent HL, not by any HL sequence. An unaccented AP (with phrasal H + boundary L) does NOT trigger catathesis (§3.1).

structure BeckmanPierrehumbert1986.IntermediatePhrase : Type

An intermediate phrase: a sequence of accentual phrases terminated by a phrase accent.

@cite{beckman-pierrehumbert-1986} §4.1: in Japanese, the ip is the domain of catathesis. It can be as short as a single AP and seldom contains more than three. Its boundary is marked by pause or glottalization, and the phrase-final L tone provides evidence for disjuncture.

§4.3: in English, the ip is reanalyzed from @cite{pierrehumbert-1980}'s framework. The phrase accent (H or L) is terminal to the ip, while the boundary tone is terminal to the IP.

• aps : List AccentualPhrase

  Accentual phrases within this ip

• phraseAccent : Features.Prosody.PhraseAccent

  Phrase accent: terminal tone of the ip

• aps_nonempty : self.aps ≠ []

  Non-empty: an ip contains at least one AP

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instReprIntermediatePhrase : Repr IntermediatePhrase

Equations

• BeckmanPierrehumbert1986.instReprIntermediatePhrase = { reprPrec := BeckmanPierrehumbert1986.instReprIntermediatePhrase.repr }

def BeckmanPierrehumbert1986.instReprIntermediatePhrase.repr : IntermediatePhrase → ℕ → Std.Format

Equations

• One or more equations did not get rendered due to their size.

structure BeckmanPierrehumbert1986.IntonationPhrase : Type

An intonation phrase: one or more intermediate phrases terminated by a boundary tone (§4.2).

• ips : List IntermediatePhrase

  Intermediate phrases within this IP

• boundaryTone : Features.Prosody.BoundaryTone

  Boundary tone: terminal tone of the IP

• ips_nonempty : self.ips ≠ []

  Non-empty

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instReprIntonationPhrase : Repr IntonationPhrase

Equations

• BeckmanPierrehumbert1986.instReprIntonationPhrase = { reprPrec := BeckmanPierrehumbert1986.instReprIntonationPhrase.repr }

def BeckmanPierrehumbert1986.instReprIntonationPhrase.repr : IntonationPhrase → ℕ → Std.Format

Equations

• One or more equations did not get rendered due to their size.

def BeckmanPierrehumbert1986.IntonationPhrase.terminalContour (ip : IntonationPhrase) : Features.Prosody.TerminalContour

The terminal contour of an IP is composed from the phrase accent of its final ip and the IP boundary tone. This is the structural decomposition that @cite{beckman-pierrehumbert-1986} §4.2–4.3 establish and that @cite{steedman-2000} uses in the CCG Tune type.

Equations

• ip.terminalContour = { phraseAccent := (ip.ips.getLast ⋯).phraseAccent, boundaryTone := ip.boundaryTone }

def BeckmanPierrehumbert1986.ipRegisterSpecs (ip : IntermediatePhrase) : List Phonology.Autosegmental.RegisterTier.TRN

Register specs within a single ip: catathesis chains.

Equations

• BeckmanPierrehumbert1986.ipRegisterSpecs ip = BeckmanPierrehumbert1986.catathesisToRegisterSpecs ip.aps

def BeckmanPierrehumbert1986.ipRegisterSpecsAcrossIps (ips : List IntermediatePhrase) : List (List Phonology.Autosegmental.RegisterTier.TRN)

Register specs across an IP: each ip resets the register. This is the structural basis for "catathesis is blocked by ip boundaries" (§4.1, §4.4, Figs. 17–18 vs. 12–13).

Each ip gets independent register specs — the l features from one ip do not propagate into the next.

Equations

• BeckmanPierrehumbert1986.ipRegisterSpecsAcrossIps ips = List.map BeckmanPierrehumbert1986.ipRegisterSpecs ips

inductive BeckmanPierrehumbert1986.CatathesisTiming : Type

Where catathesis takes effect relative to the triggering accent.

@cite{beckman-pierrehumbert-1986} §3.3: in Japanese, catathesis applies within the accent itself (affecting the trailing L). In English, catathesis applies after the second tone of the triggering bitonal accent.

• withinAccent : CatathesisTiming
• afterAccent : CatathesisTiming

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instReprCatathesisTiming : Repr CatathesisTiming

Equations

• BeckmanPierrehumbert1986.instReprCatathesisTiming = { reprPrec := BeckmanPierrehumbert1986.instReprCatathesisTiming.repr }

def BeckmanPierrehumbert1986.instReprCatathesisTiming.repr : CatathesisTiming → ℕ → Std.Format

Equations

• One or more equations did not get rendered due to their size.

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instDecidableEqCatathesisTiming : DecidableEq CatathesisTiming

Equations

• BeckmanPierrehumbert1986.instDecidableEqCatathesisTiming x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯

structure BeckmanPierrehumbert1986.IntonationSystem : Type

Language-specific intonation system parameters.

@cite{beckman-pierrehumbert-1986} §6: Japanese and English share the same prosodic hierarchy but differ in how accents relate to the lexicon, the size of the pitch accent inventory, and whether unaccented phrases are possible.

• accentSpec : Features.Prosody.AccentSpecification

  How accents are specified (lexical vs postlexical)

• accentShapes : List Features.Prosody.PitchAccent

  The set of contrastive pitch accent shapes (excluding null)

• hasUnaccented : Bool

  Whether lexically unaccented words/phrases exist

• apBoundaryLAlwaysPresent : Bool

  Whether the accentual phrase boundary L is always present

• catathesisTiming : CatathesisTiming

  When catathesis applies relative to the triggering accent (§3.3)

def BeckmanPierrehumbert1986.instReprIntonationSystem.repr : IntonationSystem → ℕ → Std.Format

Equations

• One or more equations did not get rendered due to their size.

@[implicit_reducible]

instance BeckmanPierrehumbert1986.instReprIntonationSystem : Repr IntonationSystem

Equations

• BeckmanPierrehumbert1986.instReprIntonationSystem = { reprPrec := BeckmanPierrehumbert1986.instReprIntonationSystem.repr }

def BeckmanPierrehumbert1986.IntonationSystem.accentInventorySize (sys : IntonationSystem) : ℕ

Number of contrastive accent shapes — derived, not stipulated.

Equations

• sys.accentInventorySize = sys.accentShapes.length

def BeckmanPierrehumbert1986.japanese : IntonationSystem

Japanese intonation system (§2, §6):

• Accent location is lexical (H*+L at specified mora)
• Single accent shape
• Unaccented words exist (and are common)
• Boundary L is always present

Equations

• One or more equations did not get rendered due to their size.

def BeckmanPierrehumbert1986.english : IntonationSystem

English intonation system (§2, §6):

• Accent shape is postlexical (chosen by intonation)
• Six contrastive accent shapes
• Every content word has an accentable syllable
• AP boundary is less clearly defined

Equations

• One or more equations did not get rendered due to their size.

theorem BeckmanPierrehumbert1986.japanese_lexical : japanese.accentSpec = Features.Prosody.AccentSpecification.lexical

Japanese has lexical accent specification.

theorem BeckmanPierrehumbert1986.english_postlexical : english.accentSpec = Features.Prosody.AccentSpecification.postlexical

English has postlexical accent specification.

theorem BeckmanPierrehumbert1986.japanese_smaller_inventory : japanese.accentInventorySize < english.accentInventorySize

Japanese has fewer accent shapes than English (derived from the actual accent lists, not a stipulated number).

theorem BeckmanPierrehumbert1986.japanese_one_accent : japanese.accentInventorySize = 1

Japanese uses exactly one accent shape.

theorem BeckmanPierrehumbert1986.english_six_accents : english.accentInventorySize = 6

English uses exactly six accent shapes.

theorem BeckmanPierrehumbert1986.japanese_accents_all_bitonal : (japanese.accentShapes.all fun (x : Features.Prosody.PitchAccent) => x.isBitonal) = true

All Japanese accents are bitonal (and therefore trigger catathesis).

theorem BeckmanPierrehumbert1986.english_has_monotonal : (english.accentShapes.all fun (x : Features.Prosody.PitchAccent) => x.isBitonal) = false

Not all English accents are bitonal — H* and L* are monotonal.

def BeckmanPierrehumbert1986.accentedAP : AccentualPhrase

An accented AP (e.g., uma'i 'delicious', Fig. 6a).

Equations

• BeckmanPierrehumbert1986.accentedAP = { accent := Features.Prosody.PitchAccent.H_star_plus_L, nWords := 1 }

def BeckmanPierrehumbert1986.unaccentedAP : AccentualPhrase

An unaccented AP (e.g., mame 'beans', Fig. 6).

Equations

• BeckmanPierrehumbert1986.unaccentedAP = { accent := Features.Prosody.PitchAccent.null, nWords := 1 }

def BeckmanPierrehumbert1986.twoAP_ip : IntermediatePhrase

Two-AP ip: accented + unaccented (Figs. 3, 5). Catathesis applies once — the unaccented AP is in a lower pitch range.

Equations

• One or more equations did not get rendered due to their size.

theorem BeckmanPierrehumbert1986.twoAP_ip_specs : ipRegisterSpecs twoAP_ip = [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty]

The register specs for this ip show one downstep.

theorem BeckmanPierrehumbert1986.twoAP_ip_pitch : Phonology.Autosegmental.RegisterTier.realizePitch 4 (ipRegisterSpecs twoAP_ip) = [3, 3]

Pitch realization: from baseline 4, the accented AP is at 3 (downstepped by catathesis) and the unaccented AP stays at 3.

def BeckmanPierrehumbert1986.threeAP_ip : IntermediatePhrase

Three-AP ip: accented + accented + unaccented (staircase, Fig. 11).

Equations

• One or more equations did not get rendered due to their size.

theorem BeckmanPierrehumbert1986.threeAP_ip_specs : ipRegisterSpecs threeAP_ip = [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty]

Two downsteps produce a two-level staircase.

theorem BeckmanPierrehumbert1986.threeAP_ip_pitch : Phonology.Autosegmental.RegisterTier.realizePitch 4 (ipRegisterSpecs threeAP_ip) = [3, 2, 2]

Pitch realization: 4 → 3 → 2 → 2 (descending staircase).

def BeckmanPierrehumbert1986.split_ips : List IntermediatePhrase

Catathesis blocking: same APs split across two ips. The second ip starts at full pitch range, not at the compressed level.

Equations

• One or more equations did not get rendered due to their size.

theorem BeckmanPierrehumbert1986.split_ips_specs : ipRegisterSpecsAcrossIps split_ips = [[Phonology.Autosegmental.RegisterTier.TRN.downstep], [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty]]

The register specs are independent per ip.

theorem BeckmanPierrehumbert1986.split_ips_pitch_independent : List.map (Phonology.Autosegmental.RegisterTier.realizePitch 4 ∘ ipRegisterSpecs) split_ips = [[3], [3, 3]]

Each ip starts from its own baseline — catathesis does not leak across the ip boundary. The first ip has pitch [3] (from baseline 4); the second ip independently has [3, 3].

theorem BeckmanPierrehumbert1986.catathesis_without_boundary : Phonology.Autosegmental.RegisterTier.realizePitch 4 (catathesisToRegisterSpecs [accentedAP, accentedAP, unaccentedAP]) = [3, 2, 2]

Compare: if there were no ip boundary, all three APs would be in one catathesis chain, producing [3, 2, 2] instead of [3] + [3, 3]. The second AP is at pitch 3 (not 2) because the boundary resets the register.

def BeckmanPierrehumbert1986.ipToTune (ip : IntonationPhrase) (accent : Features.Prosody.PitchAccent) : CCG.Intonation.Tune

Construct a CCG Tune from a B&P IntonationPhrase and a pitch accent. The tune's terminal contour is the IP's terminal contour — the same phrase accent + boundary tone decomposition that @cite{steedman-2000} uses for prosodic CCG categories.

Equations

• BeckmanPierrehumbert1986.ipToTune ip accent = { accent := accent, terminal := ip.terminalContour }

theorem BeckmanPierrehumbert1986.tune_terminal_eq_bp_terminal (ip : IntonationPhrase) (accent : Features.Prosody.PitchAccent) : (ipToTune ip accent).terminal = ip.terminalContour

The terminal contour of a CCG Tune constructed from a B&P IP is exactly the B&P terminal contour — @cite{steedman-2000}'s Tune decomposition and B&P's IP decomposition produce the same TerminalContour by construction.

def BeckmanPierrehumbert1986.declarativeIP : IntonationPhrase

A declarative IP (L phrase accent, L% boundary tone).

Equations

• One or more equations did not get rendered due to their size.

def BeckmanPierrehumbert1986.continuationIP : IntonationPhrase

A continuation-rise IP (L phrase accent, H% boundary tone).

Equations

• One or more equations did not get rendered due to their size.

theorem BeckmanPierrehumbert1986.declarative_ip_matches_rheme_terminal : declarativeIP.terminalContour = CCG.Intonation.rhemeTune.terminal

A declarative IP has the same terminal contour as @cite{steedman-2000}'s rheme tune (H* L L%).

theorem BeckmanPierrehumbert1986.continuation_ip_matches_theme_terminal : continuationIP.terminalContour = CCG.Intonation.themeTune.terminal

A continuation-rise IP has the same terminal contour as @cite{steedman-2000}'s theme tune (L+H* L H%).

theorem BeckmanPierrehumbert1986.ip_dual_function : Features.Prosody.PitchAccent.H_star_plus_L.isBitonal = true ∧ Features.Prosody.PitchAccent.H_star.isBitonal = false

The ip/φ domain serves two independent functions:

1. Catathesis domain (B&P §4): register resets at ip boundaries
2. Focus-prosody domain (@cite{kratzer-selkirk-2020} §7): [FoC] = φ-Level-Head

Whether focus marking triggers catathesis depends on the accent inventory: guaranteed in Japanese (all accents are bitonal), not guaranteed in English (H* is monotonal).

theorem BeckmanPierrehumbert1986.japanese_foc_always_triggers_catathesis : (japanese.accentShapes.all fun (x : Features.Prosody.PitchAccent) => x.isBitonal) = true

In Japanese, [FoC] spellout at φ-level always triggers catathesis: the only accent shape (H*+L) is bitonal. Focus-marking and catathesis are inseparable in the Japanese system.

Proof chain: [FoC] → φ-Level-Head (K&S 34) → pitch accent (head = most prominent) → bitonal (Japanese has only H*+L) → catathesis (B&P §3).

theorem BeckmanPierrehumbert1986.english_foc_may_not_trigger_catathesis : Features.Prosody.PitchAccent.H_star.isBitonal = false

In English, [FoC] spellout at φ-level does NOT guarantee catathesis: the default rheme accent H* is monotonal, so catathesis only occurs when the speaker selects a bitonal accent shape (H*+L, H+L*, L*+H, L+H*).

theorem BeckmanPierrehumbert1986.more_catathesis_lower_pitch : List.Forall₂ (fun (x1 x2 : ℤ) => x1 ≤ x2) (Phonology.Autosegmental.RegisterTier.realizePitch 4 [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty]) (Phonology.Autosegmental.RegisterTier.realizePitch 4 [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty, Phonology.Autosegmental.RegisterTier.TRN.empty])

More catathesis (more bitonal accents) produces lower pitch: replacing a registerless AP with a bitonal-accented one produces pointwise lower-or-equal pitch realization.

theorem BeckmanPierrehumbert1986.catathesis_blocking (specs : List Phonology.Autosegmental.RegisterTier.TRN) {m n : ℤ} (h : m ≤ n) : List.Forall₂ (fun (x1 x2 : ℤ) => x1 ≤ x2) (Phonology.Autosegmental.RegisterTier.realizePitch m specs) (Phonology.Autosegmental.RegisterTier.realizePitch n specs)

Catathesis blocking: realizing an ip from the higher (reset) starting offset produces pointwise higher-or-equal pitch than continuing from a compressed offset. When an ip boundary resets the register to offset n, subsequent pitches are at least as high as if catathesis had continued from any lower offset m ≤ n.

Direct application of realizePitch_baseline_mono.

theorem BeckmanPierrehumbert1986.catathesis_blocking_concrete : List.Forall₂ (fun (x1 x2 : ℤ) => x1 ≤ x2) (Phonology.Autosegmental.RegisterTier.realizePitch 3 [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty]) (Phonology.Autosegmental.RegisterTier.realizePitch 4 [Phonology.Autosegmental.RegisterTier.TRN.downstep, Phonology.Autosegmental.RegisterTier.TRN.empty])

Concrete catathesis blocking: a fresh ip starting from offset 4 (reset) yields higher pitches [3, 3] than the same specs continued from a compressed offset 3, which give [2, 2].

theorem BeckmanPierrehumbert1986.catathesis_specs_no_raising (aps : List AccentualPhrase) (s : Phonology.Autosegmental.RegisterTier.TRN) (hs : s ∈ catathesisToRegisterSpecs aps) : s = Phonology.Autosegmental.RegisterTier.TRN.empty ∨ s = Phonology.Autosegmental.RegisterTier.TRN.downstep

Catathesis register specs only contain TRN.empty (maintain) or TRN.downstep (lower) — never TRN.upstep (raise). This ensures catathesis is monotonically compressive.

theorem BeckmanPierrehumbert1986.japanese_catathesis_timing : japanese.catathesisTiming = CatathesisTiming.withinAccent

Japanese catathesis timing: catathesis applies within the accent.

theorem BeckmanPierrehumbert1986.english_catathesis_timing : english.catathesisTiming = CatathesisTiming.afterAccent

English catathesis timing: catathesis applies after the accent.