The Japanese Consonant Inventory: Phonemes, Allophones, and the Kana Chart
The Japanese consonant inventory is small and tidy on paper: roughly fifteen core consonant phonemes, plus the moraic /N/ and /Q/.12 The kana chart spells this system out row by row. A handful of allophonic alternations explain why し, ち, つ, ふ, and ひ are not what their romaji spelling implies. Reading the chart phonemically, rather than letter by letter, turns the inventory from a list into a system.
Overview
The inventory is the set of contrastive consonant sounds in the language. The kana chart is the orthography that records them. Most rows match neatly. A few cells do not, and those cells matter most for learners trying to read a Japanese consonant chart in IPA.
What a "consonant inventory" means here
A consonant phoneme is a contrastive sound: swapping it for another changes the word. An allophone is a predictable variant of one phoneme that surfaces in a specific environment. A kana is an orthographic unit.12
These three layers do not stand in one-to-one correspondence. One kana row may contain more than one phonetic consonant (the t-row hosts [t], [tɕ], and [ts]), and one phoneme may surface as several allophones.12
PHOIBLE's Stanford Phonology Archive entry for Japanese lists roughly fourteen short consonant phonemes plus geminate counterparts, omitting the moraic /N/ and /Q/.3 Vance (2008) recognizes twenty-one phonemes, adding /j w ts tɕ (d)ʑ ɕ ɸ N Q/ to a smaller core of twelve; Smith (1980) recognizes twelve (/m p b n t d s dz r k ɡ h/).2 The disagreement is about whether to count allophonic surface segments and moraic phonemes, not about which sounds occur in the language.
For descriptive purposes, the widely taught consensus inventory has about fifteen core consonant phonemes (/p b t d k g s z h m n ɾ j w/ plus an underlying /ts/ or /tɕ/, depending on analysis), together with the two moraic phonemes /N/ and /Q/.142 This is the inventory the kana chart spells out.
The Japanese term for "consonant" is 子音 (shiin); the term for "phoneme" is 音素 (onso); the term for "allophone" is 異音 (ion).5
How the inventory pairs with the kana chart
The kana chart (五十音図 gojūonzu) is mora-based: each cell is a CV mora, or one of the moraic specials V, N, Q. Here, C means consonant and V means vowel. Each consonant phoneme heads one row of the chart, and the five vowel columns fill in the V slot.12
The CV-mora unit the chart is built on is treated in depth in the J-Compass article on mora and syllable. The present article fills the C in that CV unit.
Because most rows are phonemically uniform, the chart feels regular. The visible irregularities are not lexical exceptions: し where si would be expected, ち where ti would be expected, つ where tu would be expected, ふ where hu would be expected, and ひ where the consonant audibly differs from は, へ, and ほ.12 They are surface effects of the allophonic rules covered later in this article.
JLPT and learner relevance
The JLPT (Japanese-Language Proficiency Test) does not test phonology directly. The official can-do framework names "Reading and Listening" as the activities measured, with "Language Knowledge, such as Vocabulary and Grammar" supporting them.6 Phoneme-level pronunciation is not graded at any level from N5 through N1.
An accurate consonant inventory still matters for the JLPT, because every listening section assumes that the test-taker can identify [ɕ], [tɕ], [ts], [ɸ], [ç], and [ɾ] in running speech.61
This article is the consonant counterpart to the J-Compass article on the Japanese vowel inventory. Concept-level cross-references are deliberate; the vowel side is treated in depth there.
The phoneme inventory
Place and manner: the consensus chart
The native Japanese consonant phonemes organize across six places of articulation (bilabial, alveolar, alveolo-palatal, palatal, velar, glottal) and six manner classes (stop, fricative, affricate, nasal, tap, approximant).142
The table below shows the consensus surface inventory that teachers and learner references use. Cells that house allophones rather than underlying phonemes are annotated. Each allophonic cell is explained in the Allophony section below.142
| Manner / Place | Bilabial | Alveolar | Alveolo-palatal | Palatal | Velar | Glottal |
|---|---|---|---|---|---|---|
| Stop (voiceless) | p | t | k | |||
| Stop (voiced) | b | d | ɡ | |||
| Affricate (voiceless) | ts (allo. of /t/ before /u/) | tɕ (allo. of /t/ before /i/) | ||||
| Affricate (voiced) | dz (allo. of /z/) | dʑ (allo. of /z/ before /i/, or separate phoneme) | ||||
| Fricative (voiceless) | ɸ (allo. of /h/ before /u/) | s | ɕ (allo. of /s/ before /i/) | ç (allo. of /h/ before /i/) | h | |
| Fricative (voiced) | z | ʑ (allo. of /z/ before /i/) | ||||
| Nasal | m | n | (ŋ allo. of /g/) | |||
| Tap | ɾ | |||||
| Approximant | (w) | j |
The Help:IPA/Japanese key gives a representative kana token for each surface segment: [p] パ, [b] ば, [t] た, [d] ど, [k] く, [ɡ] が, [s] す, [z] ざ, [ɸ] ふ, [ç] ひ, [ɕ] し, [ʑ] じ, [tɕ] ち, [ts] つ, [dz] ざ, [dʑ] じ, [m] み, [n] な, [ɾ] ろ, [j] や, [w] わ, [ɰ̃] ん.7
Japanese-language teacher-training material covers the same set, with [ɲ] and [ŋ] highlighted as the standardized-test forms for the palatalized n-row and the velar nasal.5
The two moraic phonemes /N/ (the moraic nasal, written ん / ン) and /Q/ (the geminate slot, written small っ / ッ) sit outside the place-and-manner table. They pattern as syllable-final segments and have no fixed surface place of articulation in isolation; their allophones are determined by the following consonant.128 They are covered in a dedicated subsection below.
Obstruents (stops, fricatives, affricates)
The stop series is /p b t d k g/, forming three voicing pairs: bilabial /p b/, alveolar /t d/, and velar /k g/.127
The fricative series at the phonemic level is /s z h/, with surface allophones [ɕ ʑ ɸ ç] that the Allophony section unpacks.127
The affricate series is [tɕ dʑ ts dz] at the surface. Most analyses treat [tɕ] and [ts] as allophones of /t/ before /i/ and /u/, respectively. [dʑ] and [dz] sit ambiguously between allophones of /z/ and independent phonemes. The choice depends on whether the analyst writes ち as /ti/ (surface [tɕi]) or as /ci/ ([tɕi]).12
There is no native /v/ phoneme in modern standard Japanese, and the bilabial [ɸ] in ふ is not the same sound as English /f/.29 Both points are unpacked further below.
The voiced fricatives are generally pronounced as affricates at the start of an utterance. Thus /z/ surfaces as [dz] word-initially and as [z] elsewhere; /ʑ/ similarly alternates with [dʑ].72
Sonorants (nasals, tap, approximants)
The nasal phonemes are /m/ and /n/. Surface [ɲ] is the palatalized variant of /n/ before /i/ and /j/, that is, the consonant in に, にゃ, にゅ, and にょ. The Japanese teacher-training reference notes [ɲ] as the standardized-test form.52
The alveolar nasal /n/ also has a velar surface variant [ŋ] in the context of an allophone of /g/ word-medially.10 The bidakuon discussion in Good to know covers this.
Japanese /ɾ/ is an apical tap, made with the tongue tip. The prototypical and most common pronunciation is alveolar [ɾ] or postalveolar [ɾ̠]. Realizations also include the lateral approximant [l], the lateral tap [ɺ], and retroflex [ɭ], with retroflex variants particularly noted before /i/ and /j/.2 It is neither English /r/ nor English /l/ nor Spanish trilled /rr/, and the ら-row gets its own dedicated J-Compass treatment for learners.21
The approximants are /j/ (palatal, as in や ゆ よ) and /w/ (phonetically bilabial, as in わ).72
Modern standard /w/ contrasts only before /a/. The only native combination is わ. ゐ (wi) and ゑ (we) have merged with い and え, and を is read [o] except as the accusative particle.2 Loanword orthography reintroduces ウィ, ウェ, and ウォ via extended katakana. That is an orthographic extension on a separate shelf, not a reactivation of the native /w/ contrast.
In non-loanword vocabulary, /j/ and /w/ generally occur only in the sequences [ja, jɯ, jo] and [wa].2
The moraic phonemes /N/ and /Q/
/N/ is the moraic nasal, written ん in hiragana and ン in katakana. It fills a mora slot on its own and assimilates to the place of the following segment.27
The surface forms of /N/ are bilabial [m] before /p, b, m/, velar [ŋ] before /k, ɡ/, a nasalized vowel or moraic semivowel broadly transcribed as [ɰ̃] before vowels and fricatives, and uvular [ɴ] at the end of an utterance. Instrumental studies show considerable variation.27
/Q/ is the moraic obstruent or geminate slot, written small っ in hiragana and ッ in katakana. It also fills a mora of its own. Japanese phonotactics mostly restrict gemination to the voiceless obstruents /p t k s/ and their allophones. Phonemically, it is transcribed as /Q/ followed by the consonant, for example /aQka/, /iQsai/, /saQti/.2
The sokuon typically avoids geminating voiced consonants (g, z, d, b), the fricative h, nasals (n, m), liquids (r), or semi-vowels (w, y), though loanwords and stylized speech create exceptions.8
/N/ and /Q/ belong in any broad-sense consonant inventory, but their detailed allophony lies outside this article's scope. Two short examples ground the moraic slots:
日本人です。1
"I am Japanese."
The string にほんじん contains two moraic /N/ slots. The first surfaces as a nasalized vowel-like segment before じ; the second surfaces utterance-finally before です.2
切手をください。8
"A stamp, please."
きって contains a /Q/ slot at the second mora. /Q/ geminates the following /t/ in line with the voiceless-obstruent restriction.28
Allophony: where one phoneme surfaces as several sounds
Most kana rows correspond to one phoneme that surfaces as one consonant. A few do not, and the alternations are systematic rather than random. The five high-yield rules below explain almost every "irregular" cell on the chart.
Palatalization in the i-column (k, g, n, m, p, b, r)
Most consonants have phonetically palatalized counterparts, meaning they are pronounced with the tongue raised toward the hard palate. Pairs of palatalized and non-palatalized consonants contrast before the back vowels /a o u/.2
Before /i/ and the glide /j/, the consonants /k g n m p b ɾ/ surface as palatalized variants conventionally written with a superscript [ʲ]: [kʲ ɡʲ ɲ mʲ pʲ bʲ ɾʲ]. The Japanese teacher-training reference standardizes [ɲ] as the surface n-row variant in this environment.5
The yōon column of the kana chart, the small ゃ ゅ ょ digraphs (きゃ きゅ きょ, ぎゃ etc.), spells exactly these /Cj/ sequences. A yōon digraph occupies one mora.21 A きょう-style cell therefore maps to a single C-on-glide-on-V mora, not to two morae. The small-kana digraphs are the orthographic record of palatalization.
今日は授業があります。1
"I have class today."
きょ and ぎょ are palatalized [kʲo] and [ɡʲo]; each is one mora.2
/h/ splits three ways: [h] [ç] [ɸ]
/h/ is [ç] before /i/ and /j/, and [ɸ] before /u/.2 Elsewhere (before /a e o/) it surfaces as plain glottal [h].27
Wiktionary transcribes ひ as [çi] and ふ as [ɸɯ̟], with the diacritic indicating a slightly fronted /ɯ/.1112 The lexicalized form 富士山 is transcribed [ɸɯ̟ʑisã̠ɴ], confirming [ɸ] in actual use.13
[ç] is the ich-Laut of standard German, the sound in German ich. [ɸ] is a bilabial fricative produced with both lips close together, not with the teeth touching the lip. Neither is identical to English /h/ or /f/.27
Hepburn romanization uses English orthography to phonetically transcribe sounds, choosing letters that English readers will pronounce close to the actual Japanese sound.14 ふ is romanized "fu" not because [ɸ] equals English /f/, but because "fu" is the closest English-spelling approximation an English reader will recognize. The romanization describes English reading habits, not the Japanese phoneme.
人がいます。12
"There is a person."
ひと is [çito]; /h/ surfaces as [ç] before /i/.2
富士山に登ります。13
"I climb Mt. Fuji."
ふ in ふじさん is [ɸ], confirmed by the Wiktionary IPA transcription [ɸɯ̟ʑisã̠ɴ].13
母は本を読みます。2
"My mother reads books."
は in はは is [ha]; before /a/, /h/ surfaces as plain glottal [h].2
/t/ splits three ways: [t] [tɕ] [ts]
Before /a e o/, /t/ surfaces as plain alveolar [t] (た て と). Before /i/, it surfaces as the alveolo-palatal affricate [tɕ] (ち). Before /u/, it surfaces as the alveolar affricate [ts] (つ).27
Wiktionary transcribes ち as [t͡ɕi] and つ as [t͡sɨ], with [ɨ] as the central-vowel transcription; other references give [ɯ̟].15167
This is the second of the chart's irregular appearances. A single row, the t-row (た ち つ て と), hosts three distinct surface consonants ([t], [tɕ], [ts]) without changing what the writer considers the underlying consonant.12
Hepburn romanizes accordingly: ち as "chi" and つ as "tsu." The same logic writes [ɕi] (し) as shi and [tɕa] (ちゃ) as cha. The system aims at English-speaker phonetic accuracy rather than one-consonant-per-row transliteration.14
父はチームで働いています。15
"My father is working on a team."
ち is [tɕi] in both ちち and チーム.2
津波に注意してください。16
"Please be careful of tsunami."
つ is [ts] in つなみ; ち is [tɕ] in ちゅうい; し is [ɕ] in して.71715
手紙を書きます。2
"I write a letter."
て before /e/ is plain alveolar [t].2
/s/ before /i/ becomes [ɕ]
The s-row consonant /s/ surfaces as the alveolo-palatal fricative [ɕ] before /i/ (し), parallel to the /h/ and /t/ palatalization patterns.27 Wiktionary transcribes し as [ɕi]. Diachronically, original /si/ came to be pronounced as [ɕi].172
The voiced counterpart /z/ similarly surfaces as [(d)ʑ] before /i/ (the kana じ). At the start of a word, the affricate variant [dʑ] is preferred. In the middle of a word, the fricative [ʑ] is preferred.72
写真を撮ります。17
"I take a photo."
し is [ɕi] and しゃ is [ɕa]; the palatalized fricative carries through the yōon.2
自分の時間が好きです。7
"I like my own time."
Word-initial じ in じぶん is [dʑi]. Word-medial じ in じかん is [(d)ʑi]. す is plain alveolar [s].2
The Japanese r: one phoneme, many realisations
The liquid phoneme /r/ varies greatly by environment and dialect, with variants including [ɾ], [ɾ̠], [l], and [ɺ].2 The prototypical and most common pronunciation is an apical tap, either alveolar [ɾ] or postalveolar [ɾ̠].2 Retroflex variants [ɭ] are noted particularly before /i/ and /j/, consistent with the broader i-column palatalization pattern.2
The tap [ɾ] is articulated by a single quick flick of the tongue tip against the alveolar ridge. If the same gesture has lateral airflow, it yields [l]-like surface tokens. With extra contact, it yields [d]-like surface tokens. None of these is English /r/, which is a continuant approximant.21
A tap is one quick contact, not a sustained continuant and not a Spanish trill. Importing English /r/ (a continuant approximant with no contact) or English /l/ (a lateral approximant) makes the r-row across ら り る れ ろ sound recognizably foreign-accented. The closer model is a single flick of the tongue tip.21
来週日本に行きます。2
"I'm going to Japan next week."
ら is the alveolar tap [ɾa].2
連絡してください。2
"Please contact me."
れ and ら both surface as the tap [ɾ]; the /N/ in れんらく assimilates to alveolar [n] before /ɾ/.2
Voicing pairs and the gaps
The voicing pairs in the inventory are /p/–/b/, /t/–/d/, /k/–/g/, /s/–/z/, with /tɕ/–/dʑ/ and /ts/–/dz/ tracking the surface affricates.127
Dakuten ゛ is the orthographic mark for voicing: a diacritic most often used in the Japanese kana syllabaries to show that the consonant of a mora should be pronounced voiced. Handakuten ゜ is used with kana for morae pronounced with /h/ or /f/ to show that they should instead be pronounced with /p/.18
The chart leaves three notable gaps in the native inventory: no /v/, no English-style /θ/ or /ð/ ("th"), and no English-style continuant /l/. Loanwords handle these gaps with extended katakana, not with native phonemes.92 The extended-katakana mechanism is covered in Good to know.
Reading the kana chart through the inventory
Why the s-row, t-row, and h-row look irregular
The irregular cells are exactly the i-column and u-column cells of three rows:
- s-row: し is [ɕi] (allophone of /s/ before /i/).217
- t-row: ち is [tɕi] (allophone of /t/ before /i/); つ is [tsɯ] (allophone of /t/ before /u/).21516
- h-row: ひ is [çi] (allophone of /h/ before /i/); ふ is [ɸɯ] (allophone of /h/ before /u/).21112
These are not lexical exceptions or "irregular kana." They are the regular outputs of the allophonic rules applied to the i-column and u-column of each row. Phonemically, the chart is regular. Phonetically, three rows have an exceptional surface form in two of their cells.12
Hepburn romanization spells these cells with phonetically motivated digraphs ("shi," "chi," "tsu," "fu," "ji") rather than the row-uniform "si," "ti," "tu," "hu," "zi" used by Nihon-shiki and Kunrei-shiki. Hepburn was designed so that speakers unfamiliar with Japanese will generally be more accurate when pronouncing unfamiliar words romanized in the Hepburn style.14
Dakuten and handakuten as voicing operations
The two marks function as orthographic operators on the consonant of a mora.
- Dakuten on the k-row gives the g-row (か → が): /k/ → /g/.18
- Dakuten on the s-row gives the z-row (さ → ざ): /s/ → /z/, with the i-column surface [ɕ] becoming [(d)ʑ].182
- Dakuten on the t-row gives the d-row (た → だ): /t/ → /d/.18
- Dakuten on the h-row gives the b-row (は → ば): /h/ → /b/. This is etymologically motivated: /b/ is the historical voiced partner of older /p/, which became /h/ in most positions.181
- Handakuten on the h-row gives the p-row (は → ぱ): the small circle restores the historical /p/ behind the h-row's voiced partner /b/. The handakuten appears to be an innovation by Portuguese Jesuits, who first used it in the Rakuyōshū to accurately transcribe the consonant /p/ and its lenited form /f/.18
This article handles the phonology of the voicing operation. The orthographic mechanics of the marks themselves are covered in the dedicated J-Compass article on dakuten and handakuten.
Sino-Japanese and the historical layer
The present-day split of the t-row (た [ta] vs ち [tɕi] vs つ [tsɯ]) reflects historical palatalization and affrication of original _/ti/ and _/tu/, processes that took place well before the modern period.1 The same palatalization shift explains why /s/ surfaces as [ɕ] before /i/ and /h/ surfaces as [ç] before /i/ (and [ɸ] before /u/).
The h-row has a particularly layered history. Modern /h/ goes back to an earlier /p/, still visible in handakuten's circle and in the voiced partner /b/. That earlier sound lenited to /ɸ/ (still the surface form before /u/) and then to /h/ in most other environments.118 The kana chart therefore preserves three different historical layers of the same row.
A full historical reconstruction is outside this article's scope. Vance (2008) is the standard English-language reference for readers who want to go deeper.1
Good to know
Yotsugana: why じ ぢ ず づ collapse to two sounds
Yotsugana (四つ仮名, literally "four kana") are four kana, じ, ぢ, ず, and づ, that historically represented four distinct voiced morae in Japanese.19 In standard modern Japanese (Tokyo, Kansai), the four characters have merged into two sounds: じ and ぢ are both [(d)ʑi], and ず and づ are both [(d)zɯ]. ぢ and づ are no longer phonetically distinguished from じ and ず in mainstream speech.19
The Tosa dialect (Kōchi prefecture) preserves the four-way distinction. Older speakers differentiate between ji (じ) and di (ぢ), and between zu (ず) and du (づ), the so-called yotsugana.20 Some Kyushu dialects similarly retain a multi-way contrast.19
Modern orthography keeps ぢ and づ alive in two contexts: rendaku-derived spellings where the voiced morpheme historically began with ち or つ, and a small set of compounds where the same etymology is transparent.1921 Rendaku is a pronunciation change seen in the middle of some Japanese compound words. When morphemes beginning with ち or つ undergo it, the result is generally spelled with the kana ぢ/ヂ and づ/ヅ rather than the identically pronounced じ/ジ and ず/ズ.21
鼻血22
"nosebleed"
鼻 (はな) plus 血 (ち) undergoes rendaku, so 血 voices to ぢ; the compound is spelled はなぢ rather than はなじ.21
続く23
"to continue"
The second mora is づ, preserved from the earlier づ rather than respelled as ず.23
Pronouncing ふ as English /f/
ふ is [ɸɯ], a voiceless bilabial fricative produced with both lips brought close together without contact, while air is blown out. English /f/ is a labiodental fricative, produced with the upper teeth on the lower lip. The two articulations are different.211
The reliable physical cue is the candle-blowing gesture: round the lips toward each other and blow out, without letting the upper teeth touch the lower lip. The same gesture that would extinguish a candle forces the bilabial articulation and blocks the labiodental one. Hepburn romanizes ふ as "fu" because it is the closest English-spelling approximation an English reader will recognize, not because the sound is English /f/.14
The corrected form for a learner who reaches for English /f/ in 富士山:
富士山13
"Mt. Fuji"
The opening consonant is [ɸ], not [f]; the full Wiktionary transcription is [ɸɯ̟ʑisã̠ɴ].13
Pronouncing ひ with English /h/
ひ is [çi]. [ç] is the same fricative as the consonant in German "ich," articulated with the tongue close to the hard palate.212 English /h/ does not palatalize before /i/, so carrying English habits into Japanese produces an audibly foreign-accented ひ.
The contrast within the same row is immediate: は is [ha], ほ is [ho], but ひ is [çi]. The /h/-row consonant is realized as three different surface sounds in standard Japanese.
The corrected form for a learner who renders ひと with English /h/:
人12
"person"
ひ is [çi], with a hissing palatal fricative noticeably different from は [ha] or ほ [ho].2
Devoicing makes consonants land back-to-back
/i/ and /u/ devoice between voiceless consonants or before a pause in standard Japanese.2 The most visible effect for the consonant inventory is that すき "like" can sound as [sk̥i], with the /u/ inaudible. Likewise, です "is" can sound as [des], with the final /u/ inaudible, leaving voiceless consonants stacked back-to-back.2
This is why a beginner often hears su-ki as "ski" and de-su as "des." The syllabic vowel is acoustically gone, but the consonant is still there and still counts as part of a mora.2 For the consonant inventory, the takeaway is that the absence of an audible vowel does not subtract a mora from the count.
Bidakuon: the [ŋ] allophone of /g/ as a register marker
Onset [ŋ], called bidakuon (鼻濁音), is generally restricted to word-internal position, where it may occur after a vowel.10 The frequency of onset [ŋ] in Tokyo Japanese speech was falling as of 2008, and pronunciations with [ŋ] are generally less frequent among younger speakers.10
The pedagogical takeaway: an older NHK-style speaker may pronounce 鏡 kagami as [kaŋami] and 学校 gakkō as [ɡakkoː] (word-initial [ɡ] retained). Younger and more casual speakers usually keep [ɡ] in both positions. Learners do not need to produce [ŋ] to be understood, but should recognize it when listening to broadcast news and older media.10
Extended katakana for loanword phonemes
The extended katakana combinations cover sounds outside the native phoneme inventory. They appear mainly to represent sounds in words from other languages.9 Common foreign-sound digraphs outside the native inventory include ファ フィ フェ フォ for [fa fi fe fo] (extending the [ɸ] allophone of /h/ to other vowels, used in loans like ファミリー), ヴァ ヴィ ヴ ヴェ ヴォ for [va vi vu ve vo] (introducing /v/ for loans), ティ ディ for [ti di] (allowing /t/ and /d/ before /i/ outside the native [tɕi] [dʑi] pattern), トゥ ドゥ for [tu du] (allowing /t/ and /d/ before /u/ outside the native [tsɯ] [(d)zɯ] pattern), チェ シェ ジェ for palatal-affricate or fricative + /e/, and ウィ ウェ for [w] + /i e/, reviving the historically merged ゐ ゑ for transliteration purposes.9
These forms appear in loanwords and proper nouns and follow extended katakana orthography. They are not part of the native phoneme inventory, and a learner-level consonant chart for native vocabulary need not include them.92 A dedicated J-Compass article covers the extended-katakana inventory in full.
See also
- Difficult Japanese Sounds by Native Language: An L1-by-L1 Pronunciation Guide
- Japanese Pronunciation Drills: A Daily 5-Minute Protocol with Minimal Pairs, Shadowing, and Record-and-Compare
- Stress vs. Pitch: Does Japanese Have Stress?
- Why "Tokyo" Is Two Syllables in English and Four Morae in Japanese: Loanwords as a Timing Drill
- Common Romaji Mistakes That Mislead Pronunciation