Heiban (平板): The Flat Japanese Pitch-Accent Pattern
In Tokyo Japanese, the heiban pitch accent is the pattern in which a word has no high-to-low pitch drop inside it and no drop onto a following particle. It is the unaccented case that linguists call mukaku (無核, "no nucleus") and dictionaries number 0.12 It is the most common pattern in the native-noun inventory. That makes it the pattern learners meet most often, and the one most easily mistaken for "no pitch information at all."13
Overview
Heiban is the first of the four Tokyo pitch-accent patterns to learn for a reason. It is statistically dominant, mechanically the simplest, and defined by one practical test: what the pitch does on the particle that follows the word. This article fixes the terminology, walks through the LHHH contour mora by mora, ties the auditory pattern to the accent-0 label, explains the particle test, and works through three standard examples.
What "heiban" (平板) literally names
平板 (heiban) is written with the kanji for "flat" and "board" (literally "flat board"). Here it means "flat" or "level."2 Combined with 型 (kata, "type") through rendaku, it gives the standard term 平板型 (heiban-gata), the form used in teaching materials and Japanese-language phonology writing.2 The synonymous 平板式 (heiban-shiki) also appears in the literature.2
Phonologists writing in English typically call the same pattern unaccented. They contrast "accented" words, which carry an abrupt H→L fall in F0 (fundamental frequency, the acoustic measure of pitch), with unaccented words, which lack such a fall.1 The Japanese-internal name for the unaccented case is mukaku (無核, "no nucleus"), referring to the absence of an accent nucleus inside the word.1 Heiban, heiban-gata, mukaku, and unaccented all name the same pattern.
One-line definition
Heiban is the pitch pattern in which the word has no H→L drop anywhere inside it and no drop onto the following particle. The contour stays high from the second mora onward until the phrase ends, including any unaccented particle that attaches at the end.12
Kawahara (2015) puts the underlying contrast bluntly: "an accented vowel is assigned a High tone followed by a Low tone on the following vowel, resulting in an abrupt H(igh)-L(ow) fall in F0, whereas unaccented words do not show such a fall."1
Where heiban sits among the four Tokyo patterns
Tokyo Japanese sorts every word into one of four accent shapes: 頭高 (atamadaka, initial accent), 中高 (nakadaka, medial accent), 尾高 (odaka, final accent), and 平板 (heiban, unaccented).14 The first three each carry exactly one H→L drop somewhere in the word. Heiban is the only one of the four that carries none.1
This "at most one drop" property is the culminativity rule that organises the whole system: "there can be at most one accentual HL fall," in Kawahara's formulation.1 Heiban is the special case where the count is zero.
How heiban sounds: the (L)HHH…H contour
The initial low mora
Mora 1 of any multi-mora heiban word is low. This is the Tokyo initial rise (also called initial lowering or initial dissimilation): "the first two syllables in a word bear a LH tonal sequence, sometimes known as initial lowering or initial rise, unless the first syllable is accented."1
The L→H movement between mora 1 and mora 2 is not unique to heiban. It is a Tokyo-wide surface rule that applies to every non-atamadaka word.4 Counting that rise as a "drop" is the most common beginner error in heiban transcription.
One exception is worth flagging: when the first syllable contains a long vowel (as in /tookyoo/, Tokyo), the word can be pronounced HH without the initial rise.1
The high plateau
After the initial L on mora 1, every later mora bears H. The mechanism is tonal spreading: "tonally-unspecified syllables get their tonal specifications by copying the tone from the rightmost specified syllable."1 In an unaccented word, that rightmost specified tone is the H placed by the initial rise. As a result, the plateau continues unbroken to the right edge of the word.1
Kawahara's derivation for an unaccented n-syllable noun has three steps. The underlying form has no accent, so accentual tone assignment does not apply. Then the initial rise places L H on syllables 1–2, and spreading copies H rightward across every remaining syllable.1 The sci.lang.japan FAQ describes the resulting contour directly: "a flat pattern called heibanshiki (平板式) where a low mora is followed by high ones, LHHHH."5
One-mora heiban words
A one-mora heiban word receives no initial rise. The rise needs a second mora to land on; with only one mora available, the word simply surfaces as H in citation form.1
This has a sharp consequence: a single-mora heiban word is phonetically indistinguishable from a single-mora atamadaka word in isolation. Kawahara puts the disyllabic case in similar terms: "the distinction between finally-accented words (e.g., /kaki/ 'fence') and unaccented words (e.g., /kaki/ 'persimmon') are phonetically very similar, if not identical, when they appear in isolation."16 To tell them apart, you need a particle. The particle-test section below gives the full treatment.
What heiban is NOT
Heiban is not a true monotone. The LH rise between mora 1 and mora 2 is real, audible, and obligatory in multi-mora words. If you flatten it out, you get the robotic-sounding contour that L1-stress speakers often produce when they are told a word is "flat."14
Heiban is also not a tone register in the way a tone language uses tone. Japanese "lexically uses only two levels of tonal heights (High and Low, and not, for example, Mid)," and culminativity restricts every word to at most one H→L fall. The system therefore has no room for the independent per-syllable tones of Mandarin or Cantonese.1 The "flat" label refers to the absence of an internal drop, not to the absence of pitch variation.
The accent-0 label
Why heiban is numbered 0
The accent number that dictionaries print names the mora after which the H→L drop occurs. An accent-2 word drops after mora 2; an accent-4 word drops after mora 4. Heiban has no drop, so there is no mora index to print. The integer 0 is the reserved "no drop" value.72
The NHK 2016 dictionary makes the convention explicit by introducing a \ symbol for the drop point and reserving a final overline ￣ for words with no drop.7 Wiktionary's pronunciation field on the term itself equates the two: "へーばんがた [hèébáńgátá] (Heiban – [0])."2
How heiban appears in OJAD, Jisho, and NHK
OJAD (the Online Japanese Accent Dictionary), hosted by the Minematsu and Saito Laboratories at the University of Tokyo's Graduate School of Engineering, tags unaccented entries with the integer 0. It renders the pitch contour as an unbroken horizontal H run after the initial L; the bundled Suzuki-kun prosody tutor visualises the same contour for arbitrary input sentences.8
Jisho and Yomitan, which both show accent data ultimately sourced from NHK and academic dictionaries, also use [0] for heiban entries.2 The NHK 2016 dictionary itself uses the overline-without-fall-mark convention. It catalogues roughly 75,000 entries that serve as the broadcaster's reference standard.7
Reading a heiban diagram back into morae
The mapping from a heiban dictionary entry to a per-mora L/H string is fixed: mora 1 is L, every later mora is H, and any following particle is H. This is Kawahara's derivation in one sentence: no underlying accent → initial rise on σ1–σ2 → H spreads rightward.1
The three canonical worked examples appear end to end in the Worked examples section below. For decoding any other heiban entry, the recipe is the same: count the morae, mark mora 1 as L, mark the rest as H, and remember that an attached particle joins the plateau.
Behaviour under particle attachment: the heiban test
The "H sustains onto the particle" rule
When が, は, を, の, に, or で attaches to a heiban word, the particle itself is pronounced H. The plateau that started on mora 2 continues without interruption across the word boundary and onto the particle.14
The mechanism is the same tonal-spreading rule that built the plateau in the first place. Particles are themselves tonally underspecified, so they inherit the rightmost specified tone: the plateau H. This is the same copy operation that gave every mora past the first its H.1 Wikipedia phrases the rule symmetrically: "This high pitch spreads to unaccented grammatical particles that attach to the end of the word, whereas these would have a low pitch when attached to an accented word."4
Why heiban and odaka sound identical in isolation
The most common heiban-misclassification trap is the odaka pattern. An odaka word (finally accented) carries its H→L drop after its last mora, so the drop has nowhere to land until something else follows. In citation form, an odaka word is L on mora 1 and H on every later mora, exactly like a heiban word.14
Kawahara is explicit on this point: finally-accented and unaccented words "are phonetically very similar, if not identical, when they appear in isolation."16 Wikipedia, on odaka specifically: "Words with this accent are indistinguishable from accentless words unless followed by a particle such as が ga or に ni."4
The odaka pattern gets its own deep-dive article in this pillar. For now, the key fact is that heiban and odaka have the same contour in isolation, and only a following particle can separate them.
The が-attachment audit
The standard linguistic convention for displaying accent is also the check a learner can run: attach the nominative particle が and listen to the が. Kawahara's own justification for the convention names the use case: "by providing an extra syllable at the end, it allows us to make clear the distinction between final-accented words and unaccented words."1
The rule is binary:
| Word type | Final word mora | Particle が |
|---|---|---|
| Heiban (0) | H | H |
| Odaka (n) | H | L |
If が is high, the word is heiban; if が is low, the word is odaka. OJAD's Suzuki-kun renders this visually: the heiban-plus-particle contour is an unbroken horizontal line, while the odaka-plus-particle contour shows the H step ending at the word edge and the particle dropped to L.8
Here is a side-by-side example. 友達 (tomodachi, heiban) + が → と(L) も(H) だ(H) ち(H) が(H), an unbroken plateau across the boundary. 妹 (imōto, accent 4, odaka in its four-mora reading) + が → い(L) も(H) う(H) と(H) が(L), with the drop appearing on the particle.14
Attach が, listen to the が: high が is heiban, low が is odaka. Nothing else in the contour matters for this specific distinction, because everything before the が is identical in the two patterns.1
Limits of the test
The が-attachment audit only distinguishes heiban from odaka. The other two patterns reveal themselves earlier. Atamadaka drops between mora 1 and mora 2, producing a HL contour that sounds different from heiban from the very start of the word. Nakadaka drops somewhere in the middle of the word, again before the particle is reached.1 By the time the が arrives, those two have already given themselves away.
The test is therefore a heiban-versus-odaka boundary tool, not a general accent identifier. For a general decoding workflow, and for minimal-pair drills on the heiban/odaka overlap, see the dedicated minimal-pairs article in this pillar.
Why heiban is the statistically dominant pattern
More than half of native nouns are heiban
In Kubozono's corpus database, 71% of the 2,220 native (yamato-kotoba) nouns surveyed are unaccented, that is, heiban.13 The same database reports heiban accent for 51% of 4,939 Sino-Japanese (kango) nouns, and for only 7% of 778 loanwords.13
The careful public statement of the claim is that heiban is the most common pattern in the native-noun inventory: more than half, by a comfortable margin, and in fact closer to three-quarters in the Kubozono corpus. For Sino-Japanese nouns, heiban is also the majority pattern, but by a much narrower margin.13
Native vs. Sino-Japanese vs. loanword skew
The same corpus shows almost the opposite pattern for loanwords. Kawahara summarises Kubozono's finding directly: "93% of the loanwords in his corpus (N = 778) are accented, whereas only 29% of the native words (N = 2,220) are accented."13 Loanwords skew strongly accented; native words skew strongly unaccented; Sino-Japanese words sit roughly at the midpoint.
The default loanword accent location is the antepenultimate mora (the third mora from the end of the word). McCawley (1968) and the later Japanese-phonology literature treat this as a default accent assignment rule.1 Kurisumasu (Christmas) and Arachia (Appalachia) both surface with antepenultimate accent under this rule, not as heiban.1
The practical consequence for a learner is that katakana words are a poor place to assume heiban. Even though heiban is the dominant pattern overall, the loanword sub-inventory leans the opposite way.13
Heiban as the verb-class default
For verbs and adjectives, the accent contrast collapses to a single binary: "Verbs and adjectives do not contrast in terms of the location of accent; rather, the contrast is simply a matter of accented vs. unaccented."1 A verb is either unaccented (heiban) or accented. If it is accented, the accent location is predictable from the inflection class.
For unaccented (heiban-class) verbs, "most forms are unaccented except when one of the two dominant suffixes is attached (the polite form and the volitional form)."1 The dictionary form, the plain past, the te-form, and most other inflections of a heiban verb all surface unaccented. Certain dominant suffixes inject an accent in specific cells. The full conjugation paradigm for accented and unaccented verbs belongs in its own pillar article; for present purposes, the key point is that heiban is the default for verb stems.
For i-adjectives, the dictionary-form distribution still tilts toward heiban. The wrinkle is that "in recent years, unaccented adjectives are becoming accented, especially among young speakers, which results in the neutralisation of the accentual contrast in adjectives."1 The contrast is slowly weakening for adjectives, but dictionary-form heiban entries remain the standard.
Implication for guessing
Read as a frequency-based nudge, not as a teaching strategy, the corpus numbers say something concrete. For a native noun of unknown accent, the prior probability of heiban is roughly 0.71; for a Sino-Japanese noun, roughly 0.51; for a loanword, roughly 0.07.13 A guess weighted by those priors will beat a uniform guess. But for any single word a learner actually needs to use, the proper move is to look it up.
Worked examples
さくら (sakura, 0)
桜 (sakura) is a three-mora native noun, accent number 0. The per-mora contour is さ(L) く(H) ら(H). With the nominative particle が attached, the contour extends to さくらが LHHH, with が joining the plateau. Kawahara attests /sakura/ as the unaccented base for the derived personal name sa'kura, which surfaces with antepenultimate accent. This cleanly illustrates how heiban can serve as the underlying form for accented derivatives.1
がっこう (gakkō, 0)
学校 (gakkō) is a four-mora Sino-Japanese compound, accent number 0. The four morae are が, the sokuon っ, こ, and the long-vowel mora う. Both the sokuon and the second half of a long vowel count as full morae in Tokyo Japanese.1 The per-mora contour is が(L) っ(H) こ(H) う(H). With the topic particle は attached, the contour is がっこうは LHHHH.
The mora count holds because, in Kawahara's summary, "in Japanese, any vowel, a coda nasal, and the second half of a geminate are moraic."1 A separate pillar article on the mora-versus-syllable distinction works through this counting in detail. For now, がっこう demonstrates that the heiban recipe applies to special morae (sokuon, long-vowel second half) the same way it applies to ordinary CV morae.
The compound 高等学校 (kōtō-gakkō, "high school") attests がっこう's heiban status indirectly: in the compound, accent lands on the が of がっこう by the Tokyo compound-accent rule, which presupposes that the second noun contributes its unaccented stem to the compound.4
ともだち (tomodachi, 0)
友達 (tomodachi) is a four-mora native compound, accent number 0. The per-mora contour is と(L) も(H) だ(H) ち(H). With the comitative particle と attached, the contour is ともだちと LHHHH. Kawahara records tomodati as unaccented in the compound onna+tomodati → onna-to'modati ("female friend"). In that compound, the compound-accent rule reassigns accent to the second noun's initial mora precisely because the second noun is itself heiban in isolation.1
Reading the three together
Every heiban word, regardless of mora count, follows the same recipe: L on mora 1, H on every later mora, and H on the following particle. This is the article's takeaway in one line. It is also Kawahara's tonal-spreading derivation compressed into a single rule.1
| Word | Morae | Per-mora pitch | With particle |
|---|---|---|---|
| さくら | 3 | さ(L) く(H) ら(H) | さくらが: LHHH |
| がっこう | 4 | が(L) っ(H) こ(H) う(H) | がっこうは: LHHHH |
| ともだち | 4 | と(L) も(H) だ(H) ち(H) | ともだちと: LHHHH |
Good to know
Heiban is not "no pitch"
The audible LH rise between mora 1 and mora 2 is real and obligatory in any multi-mora heiban word. Kawahara's initial-rise rule states that "the first two syllables in a word bear a LH tonal sequence... unless the first syllable is accented."1 Learners coming from a stress-accent L1 sometimes hear "flat" or "no accent" and produce a literal monotone, flattening the rise into a single sustained pitch. That contour sounds robotic and is not heiban. It is a pronunciation mistake. The article on stress-versus-pitch perception in this pillar walks through the perceptual root cause.
One-mora heiban words need a particle to be identifiable
A one-mora noun like 木 (ki, "tree") surfaces as H in citation form under either a heiban or an atamadaka analysis. The initial-rise rule needs two morae to operate on, and a one-mora word never reaches that environment.1 The pattern is recoverable only from the pitch of the following particle: 木が pronounced with H が is heiban; with L が, atamadaka. This is why monomoraic words are over-represented in heiban-versus-atamadaka confusion. It is also why dictionaries that print accent numbers for one-mora words rely on the convention rather than on any in-isolation acoustic distinction.
Loanwords are not a safe place to assume heiban
The native-noun heiban skew does not carry over to katakana words. Kubozono's corpus reports 93% of loanwords as accented, with only 7% heiban. The default loanword accent location is the antepenultimate mora.13 Kurisumasu (Christmas) is accented, terebi (television) is accented, and テーブル is one of the comparatively rare three-mora heiban exceptions, not the rule.1 A learner who applies "native nouns skew heiban" to loanwords will be wrong roughly nine times out of ten. The deeper landscape of loanword accent is treated in the pitch-accent overview in this pillar.
Heiban behaviour described here is Tokyo-specific
The LHHH contour and the H-sustains-onto-the-particle rule describe NHK / Tokyo standard Japanese. Kawahara is explicit that "although there is a wealth of literature on the accent patterns of non-Tokyo dialects, it is far beyond the scope of this chapter to discuss them."1 Keihan (Kyoto–Osaka) realisations of the same dictionary-heiban word can begin H or fall earlier in the word. Tōhoku and other regional systems diverge further. A dedicated article on regional variation in this pillar handles the divergences. For the Tokyo-standard reader, the recipe in this article is the one to internalise first.
The 0 in OJAD is not the same as silence
A recurring beginner confusion is to read accent 0 as "no pitch information," as though the dictionary were declining to specify a contour. It is not. 0 marks the absence of an internal H→L drop. The contour is fully specified by the heiban derivation: L on mora 1, H on every later mora, H on any attached particle.78 The integer 0 records the absence of a transition, not the absence of pitch.
See also
- Should You Learn Pitch Accent? An Honest Cost-Benefit Analysis
- Rendaku: When K Becomes G in Compound Words
- Long vs. Short Vowels in Japanese: The Distinction Beginners Miss
- Japanese Pronunciation Drills: A Daily 5-Minute Protocol with Minimal Pairs, Shadowing, and Record-and-Compare