Skip to main content

How to Predict the Reading of an Unknown Kanji Compound: The On+On Default, Jūbako, Yutō, and the Look-It-Up Bucket

Predicting the reading of a kanji compound is less a single rule than a short decision flow. Assume on+on, check a handful of named exceptions in a fixed order, and recognize the small set of compounds where prediction stops working and the dictionary takes over.12 If you run the flow on every unfamiliar two-kanji string, you train the right intuition even when you miss. Each branch you consider is itself a piece of structural knowledge.3

Overview

Japanese two-kanji compounds sort into four reading patterns, plus one off-flowchart exit. The four patterns are on + on, kun + kun, on + kun (called jūbako-yomi 重箱読み), and kun + on (called yutō-yomi 湯桶読み).24 The fifth route, jukujikun (熟字訓), gives the whole compound a reading that shares no segment with any of the kanji's normal readings. It cannot be derived by rule.56

The on+on pattern is the dominant default, reported in the 75–80% range across corpus and pedagogy literature.178 The two mixed patterns are minority routes named after their archetypal examples. Both archetypes are autological, meaning they demonstrate the pattern they name: 重箱 itself reads on+kun, and 湯桶 itself reads kun+on.24

The two archetypes name themselves

The order of the syllables in the pattern's name matches the order of the readings: jūbako = on () + kun (bako), yutō = kun (yu) + on (). Once that hook lands, the two patterns are easier to keep apart.24

The decision in one minute

The four-pattern landscape

Together, the four patterns and the jukujikun exit form a five-way classification for any two-kanji compound a learner is likely to meet. The named patterns are well established in standard reference dictionaries.24

The on+on share rises in technical, legal, and news writing, where kango is more common. It falls in literary or everyday prose, where wago and mixed compounds are more common.19 The Sino-Japanese (kango) and native (wago) layers together account for roughly 88–92% of vocabulary across all BCCWJ (Balanced Corpus of Contemporary Written Japanese) registers. Foreign loans (gairaigo) take the rest.910

The most rigorous public source for the per-pattern distribution is kanjidatabase.com. It is built on the Mainichi Newspaper corpus 2000–2010, with approximately 282.8 million morpheme tokens after excluding proper nouns.7 The exact percentage depends on whether the count is by type (distinct compounds) or by token (running occurrences).7

Why a procedure beats memorization

Two-kanji compounds form a productive open class in Japanese. New ones enter the lexicon constantly through news, academia, and trade names, so a learner cannot memorize the whole inventory.13

A staged procedure, meaning a default plus exception checks, trains the right intuition even when the guess is wrong. Each branch the learner considers is itself a piece of structural knowledge.3 The goal is a confident first guess plus a clean rule for when to stop guessing and consult a dictionary.

Step 1: default to on+on

Why on+on dominates two-kanji compounds

Most Japanese two-kanji compounds entered the language as kango (Sino-Japanese vocabulary) borrowed from Middle Chinese between the 5th and 9th centuries. They arrived as whole compound units, with their Chinese-derived on-readings already locked in.11112

This historical fact, not a synchronic rule (a rule operating in the language now), is what makes the on+on pattern statistically dominant. The same productive process still operates in coinages like 携帯 keitai, 環境 kankyō, and 情報 jōhō.112

Reports of the on+on share cluster around 75–80% of two-kanji compounds. The exact figure depends on the corpus (technical writing higher, literary writing lower) and on whether the count is by type or by token.178

学校がっこう13
"school"

電車でんしゃ3
"electric train"

図書館としょかん3
"library"

経済けいざい3
"economy"

What on+on looks like in practice

In a monolingual or learner dictionary, on'yomi are conventionally written in katakana while kun'yomi are written in hiragana. A compound whose components both appear with katakana readings in the entry is almost always read on+on.3

Having no okurigana on either kanji is necessary, but not sufficient, for on+on. An okurigana tail almost always signals a kun reading.8

If you need the foundation before the flow chart, review the deep dive on the two-reading system. It explains why on+on, kun+kun, and mixed patterns exist in the first place.

Step 2: pick the right on-reading when several exist

Domain signals: Buddhist, classical, scientific

Japanese on'yomi fall into three main historical layers, plus a fourth catch-all. The same kanji often carries readings from more than one layer.1114

  • Go-on (呉音): the oldest layer, borrowed from the Wu region of southern China between the 5th and 6th centuries; preserved heavily in Buddhist vocabulary.1114
  • Kan-on (漢音): borrowed from the standard pronunciation of Chang'an during the Tang dynasty (7th–9th centuries); the dominant layer in modern compound coinage.1114
  • Tō-on (唐音), also called Sō-on: later borrowings from the Song and Ming dynasties (from the 10th century onward), brought mainly by Zen monks and merchants; preserved in a small set of words such as 椅子 isu and 蒲団 futon.1114
  • Kan'yō-on (慣用音): "customary" readings that drifted from any of the above and became standardized.14

A compound's layer usually signals its domain. The kanji 行 carries go-on gyō, kan-on , and tō-on an, and each surfaces in a different register.14

銀行ぎんこう3
"bank"

修行しゅぎょう3
"Buddhist ascetic training"

行脚あんぎゃ4
"pilgrimage on foot"

When the dictionary lists two on-readings, prefer the kan-on default

Kan-on is the modal, or most common, layer for modern jukugo. Among kanji with multiple on-readings, the kan-on entry is more often the one a learner will meet first in everyday compounds.14 It is not a strict majority, and no per-corpus count is available. So kan-on functions as a confident first guess rather than a guarantee.

When a kanji's dictionary entry lists multiple on'yomi and the compound is not flagged as Buddhist or culinary, kan-on is the safer first guess.14 Buddhist vocabulary (経 kyō "sutra" rather than kei; 行 gyō; 明 myō) and a small set of food and household items (布団 futon, 行灯 andon) carry go-on or tō-on, respectively.11144

Tō-on words mark Zen-and-trade contact dates

椅子 isu, 蒲団 futon, 行脚 angya all carry tō-on readings, marking them as Song- or Ming-period borrowings tied to Zen Buddhism and continental trade. Their borrowing dates are encoded in the reading layer itself.1114

The historical layers above earn their own treatment in the dedicated article on the Sino-Japanese reading strata.

Step 3: use the right kanji's phonetic component (形声)

How phonetic series predict the on-reading

Keisei (形声) characters are phono-semantic compounds and one of the traditional 六書 categories. They pair a semantic element ("radical") with a phonetic element that signals the character's on-reading.215 In Shirakawa's classification, 形声 characters account for 61% of the 2,136 jōyō kanji overall. That breaks down as 49% of the 1,026 educational (kyōiku) kanji and 72% of the remaining 1,110 more advanced jōyō characters.16 Pedagogy sources report broader estimates of 70–90% depending on the inventory and classification scheme.15

The 青 sei ("blue/green") series is one of the most useful series for learners. The on-reading sei (sometimes the alternate shō) surfaces across compounds that contain 青 on the right-hand or enclosed side.154

KanjiComposition (semantic + phonetic)On'yomiVerified compound
氵(water) + 青sei, shō清潔 seiketsu
日(sun) + 青sei晴天 seiten
米(rice) + 青sei, shō精神 seishin
言(speech) + 青sei, shin申請 shinsei
青 + 争sei, 静止 seishi
忄(heart) + 青, sei感情 kanjō

清潔せいけつ3
"cleanliness, hygiene"

晴天せいてん3
"clear weather"

When the phonetic component lies

Sounds changed over the centuries between the original Middle Chinese borrowing and modern Sino-Japanese. As a result, a given phonetic element can carry different on-readings in different characters. The learner verifies, never trusts blindly.1115

A clean counter-example is 寺 ji as a phonetic component. The element predicts ji inside 持 ji ("hold") and 時 ji ("time"), but inside 詩 it shifts to shi ("poem").15 The first guess is still worth making. The phonetic match is evidence, not proof, and a quick verification step catches the divergent cases.

Hamilton's rule of thumb is that the phonetic gives a "first guess" with an estimated 50–70% success rate on jōyō kanji, depending on the series.15 Treat a phonetic match as a strong lead that survives or fails on the second check.

The phonetic check is a lead, not a verdict

A 50–70% success rate means the phonetic component is right about as often as a coin flip weighted in the learner's favor. Always confirm the predicted on-reading against the kanji's dictionary entry or against a compound the learner already knows. Never overwrite a known reading on the strength of a phonetic match alone.15

Cross-link to the six-categories article

The 形声 category and the other five categories are covered in the dedicated article on the traditional 六書 typology.

Step 4: check for rendaku on the second element

When rendaku fires

Rendaku (連濁, "sequential voicing") is a sound change in compounds. A voiceless initial consonant of the second element becomes voiced (k → g, s → z, t → d, h → b).1718

Rendaku applies productively to native (wago) second elements. In practice, that means mostly kun+kun compounds and the jūbako (on+kun) pattern, when the kun-read second element starts with a voiceless obstruent.1718

Sino-Japanese (kango) second elements do not undergo rendaku, which is why pure on+on compounds keep the second element's underlying voiceless onset.17182

花火はなび3
"fireworks"

手紙てがみ19
"letter"

朝日あさひ3
"morning sun"

When rendaku is blocked

Four blockers cover most cases.

  • Lyman's Law: rendaku is blocked when the second element already contains a voiced obstruent (g, z, d, b) elsewhere. This is the strongest blocker identified in the literature.171820
  • Sino-Japanese morphology: kango second elements resist rendaku, which is why on+on compounds keep their voiceless onsets.1718
  • Recent loanwords: gairaigo second elements almost never undergo rendaku.18
  • Dvandva ("A and B") compounds: coordinative compounds, where the two elements are joined as equals rather than arranged as modifier and head, tend not to voice.18

The textbook contrast for dvandva blocking is 山川 yamakawa ("mountains and rivers", coordinative, no rendaku) versus 山川 yamagawa ("mountain river", modifier-head, with rendaku). The Vance and Irwin (2016) volume is the safest reference for the direction of this contrast.18

山風やまかぜ17
"mountain wind"

Lyman's Law sits inside the second element

The voiced obstruent that blocks rendaku must be inside the second element, not the first. 山風 keeps kaze voiceless-initial because kaze itself already contains /z/. 山道 yamamichi is unaffected by anything inside 山. Misreading the locus produces wrong predictions in both directions.1718

The full rendaku picture, including the productive cases and the long tail of lexical exceptions, is covered in the dedicated article on sequential voicing in kanji compounds.

Step 5: recognize the kun+kun pattern

Signals of a native compound

Kun+kun compounds (also called wago compounds) typically belong to concrete, native-feeling semantic fields: nature, body parts, daily life, and kinship.112

A short native suffix attached to a one-kanji base is a strong kun+kun signal. The endings 〜火 -bi, 〜道 -michi, 〜月 -zuki, 〜人 -hito, and 〜日 -hi all tend to take kun first elements.312

山道やまみち3
"mountain path"

月見つきみ3
"moon-viewing"

朝日あさひ3
"morning sun"

花火はなび3
"fireworks"

Why no okurigana does not mean on'yomi

The "two kanji, no kana between them, no okurigana = on+on" heuristic fails on kun+kun compounds. By definition, these compounds pack two native readings into kanji without okurigana on the second element.38

Okurigana absence is a necessary but not sufficient signal for on+on. The learner must also weigh semantic field, vocabulary register, and the presence or absence of rendaku.3

When kun+kun is also a name

Many surnames and place names are kun+kun compounds whose kanji carry their everyday native readings: 田中 Tanaka, 山田 Yamada, 中野 Nakano. The same surface compounds can also lexicalize as common nouns elsewhere.53

For proper nouns, an additional layer of irregular readings called nanori (名乗り) is sanctioned by the Jōyō Kanji List appendix and the Family-Name Kanji table. Nanori extends the kun inventory specifically for names.5

Cross-link to the names article

Proper-noun readings, including the nanori inventory and the place-name conventions that overlap with it, are covered in the dedicated article on name-only kanji readings.

Step 6: recognize the two mixed patterns

Jūbako (on + kun)

Jūbako-yomi (重箱読み) names the pattern in which the first kanji uses its on'yomi and the second uses its kun'yomi.221422 重箱 itself is the archetype that names itself: 重 is on'yomi, and 箱 (hakobako with rendaku) is kun'yomi.422

重箱じゅうばこ22
"tiered lacquer food box"

台所だいどころ234
"kitchen"

本屋ほんや234
"bookshop"

役場やくば4
"town office"

額縁がくぶち4
"picture frame"

Yutō (kun + on)

Yutō-yomi (湯桶読み) names the mirror pattern: kun'yomi first, on'yomi second.22142425 湯桶 itself is the archetype that names itself: 湯 yu is kun'yomi, and 桶 is on'yomi.214

Yutō-yomi is the less frequent of the two mixed patterns. Jūbako outnumbers yutō in standard reference dictionaries' worked lists. This reflects the historical asymmetry in which Sino-Japanese first elements naturalized more readily as modifiers than as heads.124

湯桶ゆとう214
"wooden hot-water pitcher"

場所ばしょ24
"place, location"

手本てほん264
"model, exemplar"

見本みほん264
"sample, specimen"

夕刊ゆうかん25
"evening newspaper edition"

How to detect a mixed reading before you confirm it

Three structural clues can signal a mixed pattern before the learner reaches a dictionary.

A concrete, native-feeling first kanji (湯, 場, 手, 見, 夕) paired with an abstract Sino-Japanese head is a yutō clue.4 A Sino-Japanese first element with a homely native second (重箱, 台所, 本屋) is a jūbako clue.4 More generally, a register mismatch between the two halves is the first signal that the compound is not pure on+on or kun+kun: kango formality on one side, wago concreteness on the other.14

Step 7: stop guessing when it is jukujikun

Signals that the compound is jukujikun

A jukujikun (熟字訓) is a reading assigned to a kanji compound as a whole. The reading shares no segment with the individual kanji's standard on or kun readings.5627

The giveaway is a mismatch in surface shape. The spoken form is a short, native-sounding word, but the written form is two or three Chinese-looking characters.56

大人おとな27
"adult"

今日きょう6
"today"

田舎いなか5
"the countryside"

紅葉もみじ5
"maple leaf; autumn-colored leaves"

明日あした5
"tomorrow"

Why the look-it-up branch is the right call

Jukujikun readings are stored with the compound as whole-word readings. They are not synthesized from its parts. The Jōyō Kanji List appendix of 116 jukujikun entries (常用漢字表 付表) is the authoritative finite inventory of the modern standard set. It exists precisely because these readings cannot be derived by rule.5

Spending more than a few seconds trying to predict a jukujikun reading wastes time and reinforces wrong intuitions. The correct branch is to recognize the off-flowchart status and consult a dictionary.53

The full inventory of jukujikun readings and the appendix's role in fixing the modern standard set are covered in the dedicated article on whole-word kanji readings.

The ateji boundary

Ateji (当て字) and jukujikun both sit outside the four-pattern flow chart. They break predictability for opposite reasons.284

Ateji chooses kanji for sound with disregard for meaning. The kanji's readings are honored, but the semantic match is loose or absent. Examples include 寿司 sushi (the kanji mean "long life" and "administer" but are picked for their sounds) and 珈琲 kōhī ("coffee," purely phonetic).284

Jukujikun chooses kanji for meaning with disregard for sound. The semantic match is exact (大人 = "big person" = adult), but the reading is the native word, not the sum of the kanji's readings.628

Ateji often appears in foreign loans, brand names, and pre-modern phonetic transcriptions, while jukujikun is concentrated in core native vocabulary listed in the Jōyō appendix.5284

The phonetic-kanji-selection process and its conventional inventory are covered in the dedicated article on sound-first kanji choice.

The full flow chart, in one place

The seven-step procedure, condensed

Here is the full procedure, in the order a learner should run it on an unfamiliar two-kanji compound.

The flow is iterative, not one-shot. A learner often enters at Step 1, detects a register or semantic mismatch at Step 5 or 6, and loops back through Step 4 to reapply rendaku on the now-reclassified compound.3

How to use it on a real page

Two walk-throughs show how that iteration works.

朝食 chōshoku ("breakfast"). Step 1 hypothesis: use the on+on default and predict on-readings for both kanji.13 Step 2 check: 朝 has chō (kan-on), and 食 has shoku (kan-on). The kan-on default holds.13

Step 3 check: 食 contains the phonetic 食, and shoku is consistent.13 Steps 4–7 produce no detour. Read it as chōshoku; the dictionary verifies.13

朝食ちょうしょく13
"breakfast"

手紙 tegami ("letter"). Step 1 hypothesis: the on+on default would predict shushi (手 shu + 紙 shi), a non-word a native speaker would not recognize.19 The detour starts immediately because the meaning belongs to daily life. Drop to Step 5 and reclassify it as kun+kun, te + kami.19

Step 4, the rendaku check, applies to the wago second element (k → g), giving tegami.1719 The dictionary verifies tegami.19

手紙てがみ19
"letter"

The walk-throughs show the iterative shape of the flow. Each branch supplies evidence for or against the current guess, and revising the hypothesis costs nothing because the earlier checks are still valid.3

Good to know

The 75–80 percent figure is a guide, not a guarantee

The widely cited 75–80% on+on share for two-kanji compounds is a corpus average, not a promise. Technical, legal, and news writing skew higher because they use more kango. Literary prose and conversation skew lower because they use more wago and mixed compounds.19710 A learner who reads manga or fiction will hit kun+kun and mixed compounds far more often than one who is studying for a JLPT vocabulary list. Calibrate expectations to the input source.9

Three-kanji and four-kanji compounds change the math

Three-kanji compounds usually parse as a 2+1 or 1+2 morphological structure. The four-pattern flow applies to each chunk independently. 図書館 parses as (図書) + 館, all on; 大食漢 parses as (大食) + 漢, all on.13

Four-character compounds called yojijukugo (四字熟語) almost always read on+on+on+on with classical Chinese word order. That is because they are usually direct quotations or paraphrases from Classical Chinese sources.112

一石二鳥いっせきにちょう3
"kill two birds with one stone"

Place names and surnames are their own ecosystem

Place-name (chimei) and surname (myōji) readings draw heavily on nanori, a category of kanji readings sanctioned only for proper nouns. They also draw on historical reading conventions that pre-date the modern on/kun reform.53

Examples include 中野 Nakano, 新宿 Shinjuku, 大阪 Ōsaka, 神戸 Kōbe, and 日本橋 Nihonbashi (kun+kun+on with rendaku).3 The four-pattern flow chart is built on the open-class common-noun lexicon and will misfire on proper nouns. Treat names as their own lookup category.53

Reading an on+on compound with kun'yomi because the kanji are common

A learner who treats both kanji in 山道 as on'yomi just because the compound has no okurigana will produce the non-existent reading sandō. The correct reading is yamamichi (kun+kun, with concrete native semantics overriding the on+on default). The "no okurigana → on'yomi" heuristic is a one-way implication, not a biconditional. Kun+kun compounds are a recognized exception class.38

山道やまみち3
"mountain path"

Forgetting that on+on blocks rendaku

A learner who reaches for rendaku on every compound will incorrectly predict 学校 as a voiced-second-element form. The correct reading is gakkō, with no rendaku. Rendaku is a wago-specific process, and the kango second element 校 resists it.1718

学校がっこう13
"school"

Trying to predict a jukujikun reading

A learner who reads 大人 as daijin, tainin, or ōhito is applying the four-pattern flow to a compound outside that flow. The correct reading is otona, a jukujikun lexicalized to the whole compound. The reading is not synthesizable from the parts, and the compound belongs in the look-it-up bucket.5627

大人おとな27
"adult"

Treating ateji as on+on

A learner who reads 寿司 as jushi or kotobuki-tsukasa is honoring the kanji's standard readings in a compound where the kanji were chosen for sound, not meaning. The correct reading is sushi, an ateji form: phonetic kanji selection for an existing native or borrowed word. The standard on/kun readings of the chosen kanji are bypassed.284

寿司すし4
"sushi"

The 75–80 percent figure as a confidence anchor

A useful default stance for an unfamiliar compound is this: assume on+on, kan-on, and no rendaku. The learner will be right roughly four times out of five on bare two-kanji compounds in a news article. The misses are concentrated in semantically native or register-mismatched compounds that flag themselves at Step 5 or Step 6.178 Anchoring the default keeps the learner from overfitting to the exception classes and freezing on every compound.

Pure on+on compounds carry a kango formality register

Kango is the prestige layer of the Japanese lexicon and the layer often used in formal writing. Substituting a wago synonym shifts the register noticeably. 食事 shokuji (kango) reads as neutral or formal "meal", while 食べ物 tabemono (wago) reads as everyday "food".112 The on+on / kun+kun split is not only a sound prediction. It is also a register signal.

Why jūbako outnumbers yutō

Sino-Japanese morphemes were borrowed predominantly as modifiers, meaning left-hand elements that modify native heads. Because Japanese is head-final, the kango modifier + wago head order fits the language's structure, producing jūbako. The mirror order (wago modifier + kango head) is structurally available but historically rarer. That is why yutō remains the minority pattern of the two mixed readings.14

The Jōyō appendix is a finite inventory of the standard jukujikun set

The 2010-revised 常用漢字表 lists 116 jukujikun and ateji forms in its appendix (付表). This is the closed list that a JLPT learner is expected to know on sight. The rest are learned ad hoc as the learner encounters them.5 Knowing that the appendix exists, and that it is finite, makes the look-it-up branch feel less open-ended.

Why your first guess is worth making even when wrong

Active prediction followed by verification produces stronger memory traces than passive lookup, even when the prediction is wrong. The prediction error itself is informative. A wrong guess identifies which branch of the flow chart the learner's intuition misfired on.3 The pedagogy literature on the testing effect (retrieval practice) supports this framing for vocabulary learning generally. The kanji-compound case is a domain-specific application.3

See also

References

Footnotes

  1. Shibatani, Masayoshi. The Languages of Japan. Cambridge Language Surveys. Cambridge University Press, 1990. Chapters on the Japanese lexicon and word formation. ISBN 978-0521369183. 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

  2. Kanji. Wikipedia (English), survey section on reading patterns including jūbako-yomi and yutō-yomi. https://en.wikipedia.org/wiki/Kanji 2 3 4 5 6 7 8 9 10

  3. Halpern, Jack (ed.). The Kodansha Kanji Learner's Dictionary. Kodansha International, revised expanded edition, 2013. 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

  4. 大辞林 第四版. Sanseidō, 2019. Entries: 重箱読み, 湯桶読み, 熟字訓, 当て字, 連濁. 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

  5. 文化庁 (Agency for Cultural Affairs). 『常用漢字表』 (Jōyō Kanji List), 2010 revision, including 付表 (appendix) of jukujikun and ateji readings. https://www.bunka.go.jp/kokugo_nihongo/sisaku/joho/joho/kijun/naikaku/kanji/ 2 3 4 5 6 7 8 9 10 11 12 13 14 15

  6. Jukujikun (熟字訓). English Wiktionary. https://en.wiktionary.org/wiki/%E7%86%9F%E5%AD%97%E8%A8%93 2 3 4 5 6

  7. Tamaoka, Katsuo, Shogo Makioka, Sander Sanders, and Rinus G. Verdonschot. www.kanjidatabase.com (interactive online database derived from the Mainichi Newspaper corpus 2000–2010, approximately 282.8 million morpheme tokens after excluding proper nouns). http://www.kanjidatabase.com/ 2 3 4 5 6

  8. Tofugu staff. "On'yomi And Kun'yomi in Kanji: What's the Difference?" tofugu.com. https://www.tofugu.com/japanese/onyomi-kunyomi/ (limitation: pedagogy blog; cited only for general heuristic framing). 2 3 4 5 6

  9. 国立国語研究所 (NINJAL). 『現代日本語書き言葉均衡コーパス』 (Balanced Corpus of Contemporary Written Japanese, BCCWJ), version 1.1, 2011. https://clrd.ninjal.ac.jp/bccwj/en/ 2 3 4

  10. Joyce, Terry, Bor Horoszek, and Yuu Nishina. "BCCWJ-based kanji frequency and coverage lists." Tama University, 2012. Reproduced in: National Institute for Japanese Language and Linguistics (NINJAL), BCCWJ word-frequency tables. https://pj.ninjal.ac.jp/corpus_center/bccwj/en/freq-list.html 2

  11. Frellesvig, Bjarke. A History of the Japanese Language. Cambridge University Press, 2010. Sections on the Sino-Japanese reading layers go-on, kan-on, tō-on. 2 3 4 5 6 7 8

  12. Kindaichi, Haruhiko. The Japanese Language. Translated by Umeyo Hirano. Tuttle, 1978. Section on Sino-Japanese vocabulary stratification. 2 3 4 5 6

  13. Wiktionary entry 朝食. https://en.wiktionary.org/wiki/%E6%9C%9D%E9%A3%9F 2 3 4 5 6 7

  14. On'yomi. Wikipedia (English), citing Frellesvig (2010) and Miller (1967). https://en.wikipedia.org/wiki/On%27yomi 2 3 4 5 6 7 8 9 10

  15. Hamilton, Natalie. The Kanji Code: See the Sounds with Phonetic Components and Visual Patterns. Selfish Publishing, 2018. 2 3 4 5 6 7

  16. 白川, 静. 『常用字解』. 平凡社, 2004 (Shirakawa Shizuka, Jōyō Jikai, Heibonsha, 2004). Source for the formation-type classification of jōyō kanji.

  17. Vance, Timothy J. The Sounds of Japanese. Cambridge University Press, 2008. Chapter on rendaku (sequential voicing). 2 3 4 5 6 7 8 9

  18. Vance, Timothy J., and Mark Irwin (eds.). Sequential Voicing in Japanese: Papers from the NINJAL Rendaku Project. Studies in Language Companion Series 176. John Benjamins, 2016. 2 3 4 5 6 7 8 9 10

  19. Wiktionary entry 手紙. https://en.wiktionary.org/wiki/%E6%89%8B%E7%B4%99 2 3 4 5 6

  20. White-Sustaíta, Jessica. "Lyman's Law in present-day Japanese: An OT analysis." Texas Linguistic Forum 49: 233–242, 2005.

  21. Bullock, Ben. "Can a word have mixed on and kun readings?" sci.lang.japan FAQ. https://www.sljfaq.org/afaq/juubako-yutou-yomi.html (limitation: hobbyist FAQ; cited only for the named patterns and the 湯桶 / 重箱 archetypes). 2 3 4

  22. Wiktionary entry 重箱. https://en.wiktionary.org/wiki/%E9%87%8D%E7%AE%B1 2 3

  23. 加納, 圭. "Jūbako yomi (重箱読み)." Kano Japanese School blog, 2021. https://blog.kano.ac/2021/03/03/jubako-yomi/ (limitation: language-school blog; cited only for the worked examples 台所, 本屋). 2

  24. Wiktionary entry 場所. https://en.wiktionary.org/wiki/%E5%A0%B4%E6%89%80 2

  25. Wiktionary entry 夕刊. https://en.wiktionary.org/wiki/%E5%A4%95%E5%88%8A 2

  26. 加納, 圭. "Yutō yomi (湯桶読み)." Kano Japanese School blog, 2021. https://blog.kano.ac/2021/03/04/yuto-yomi/ (limitation: language-school blog; cited only for the worked examples 手本, 見本). 2

  27. Wiktionary entry 大人. https://en.wiktionary.org/wiki/%E5%A4%A7%E4%BA%BA 2 3 4

  28. Ateji. Wikipedia (English). https://en.wikipedia.org/wiki/Ateji 2 3 4 5