A common discussion in Japanese-learning circles is the question of which romanization system is a better representation of Japanese. Of course, the best and most accurate representation of Japanese is using the native Japanese characters; however, there are several cases where romanization is a necessary evil, such as representing Japanese words to people incapable of reading Japanese characters who nevertheless need to be able to recognize or pronounce them (e.g. actors reading lines or tourists reading signs) or sorting in a list that also contains Latin-alphabet words.
As anybody with a passing familiarity with the language will be aware, there are several contenders, enjoying various degrees of popularity. I intend herein to examine the advantages and disadvantages of some of the more frequent examples, and suggest an alternative that combines the advantages of all of them (and a few more) whilst eliminating as many disadvantages as possible, to give as faithful a representation of the language as possible.
Probably the most prominent system in usage is Hepburn. This is the system most English-speakers will have encountered when reading Japanese. It aims to make Japanese pronunciation more accessible to English-speakers by representing sounds using digraphs parallel to those found in English; for example, the consonant [ɕ] that begins the pronunciation of ‹し› is written ‘sh’, by analogy with English ‹sh› (representing [ʃ]).
While this is convenient for recognition by English-speakers, it can already be seen to have several obvious flaws. Being based on the English orthography, it does not make much sense for speakers of other languages where the digraphs may not have the same meaning, or may not be digraphs at all; ‹sh› to a Finnish speaker simply represents the sound /s/ followed by the sound /h/.
This disadvantage cuts both ways when considering English-speakers who are attempting to pronounce Japanese accurately: the similarity with the English [ʃ] digraph can and often does result in novices pronouncing the word with [ʃ], when in fact the real sound [ɕ] is closer to a simple [s] in terms of mouth position. Learners of the language, who are attempting to understand the grammar and morphology of the language, must add an additional layer of recognition to associate ‹chi› with the ‘t’ row of the sound table; for example, a beginner may not immediately recognize that ‘tachi’ (立ち, standing) is related to the verb ‘tatsu’ (立つ, stands), as the word form is significantly different when expressed in Hepburn romanization, obscuring the otherwise obvious (when expressed in kana) morphological relationship. This also introduces issues in sorting: words beginning with ち are sorted with ‹c› in Latin-script lists, rather than grouped together with other words in the ‘t’ column of the sound tables as is natural in Japanese.
Furthermore, as a purely phonetic system, Hepburn introduces ambiguity in many cases as to which kana are used to write a word: the kana ぢ and じ are both written as ‹ji›, and 大【おお】, a prefix meaning ‘large’ or ‘great’, is written ‘ō’, the same as 王【おう】, ‘king’.
Nihon-siki is the name given to an ISO-standard romanization system originally intended to replace the Japanese writing system altogether. It is exceedingly regular, representing each kana as one (for bare vowels and the final syllabic /ɴ̩/, ん) or two letters, or ‘small kana’ pairs (拗音) as three. It is a faithful representation of the kana, and all native Japanese kana can be represented unambiguously.
This makes it a reasonably good encoding of a subset of Japanese, but as a system for foreigners to read it is severely lacking. Unlike Hepburn, it clearly indicates the relationship between words like ‘tati’ and ‘tatu’; however, it does not encode pronunciation information, and therefore no indication is made that the consonants of ‘unstable’ kana such as ‘ti’ are pronounced any differently from the rest of their row, and non-phonetic kana usage such as は for the topic-marking particle, pronounced /wa/, is not marked.
Its strict ‘one kana, two letters’ philosophy leads to increased regularity, but also severely restricts its capability to represent uncommon or foreign phoneme combinations: there is no way to represent the /ɸa/ sound of ファイル, for example, or to distinguish between the usual [tɕi] sound of ち vs. the foreign [ti] sound represented as てぃ in, for example, パーティー (‘party’).
Despite its general lack of ambiguity, Nihon-siki follows Hepburn in being unable to distinguish 大 from 王: both are written ‘ô’.
JSL romanization is the system of romanization used in the book Japanese: The Spoken Language. It is a purely phonemic system, and, in contrast to many other systems, includes the ability to mark pitch accent, discriminating between words that would be written the same even in kana, such as 酒 (‘alcohol’ or ‘rice wine’) and 鮭 (‘salmon’), both of which are written さけ in kana, but distinguished in speech by pitch accent: in Tokyo Japanese, 鮭 (‘salmon’) is pronounced with a pitch drop on the first syllable, whereas 酒 (‘alcohol’) is not. Due to the regional sensitivity of pitch accent, indicating it in text is rather hit-and-miss; however, even restricting oneself to the pitch accent of one dialect allows one to avoid a large class of ambiguities. It also aids learners' pronunciation by encouraging them to recall the pitch patterns on words when typing or reading.
JSL is strongly based on Nihon-siki, and suffers from many of the same drawbacks: it can only represent a very few uncommon phoneme combinations, the unstable sounds are not regularly indicated, and 大 and 王 are not distinct.
I propose tying together the three goals we have seen in the other systems (unambiguity of kana representation, unambiguity of pronunciation, and unambiguity of morphological relation) into a single system, avoiding as many of the pitfalls of previous systems as possible. As a sub-goal, the system aims to break kana apart into phonemes and represent each phoneme by a single letter, allowing any combination of phonemes, regardless of whether or not the combination is used in native Japanese. This increases both the regularity and the expressive power of the system.
In order to preserve a strict ‘one phoneme, one letter’ correspondence and to indicate variants without obscuring relationships to other kana, the system makes use of various diacritics available in Unicode. The application of these is regular and quite simple; however, since ease of input using a Latin-oriented keyboard layout is one of the goals of a romanization system, an ASCII encoding of the system, intended for translation to the full system and for use in legacy encoding-constrained contexts, is detailed in the appendix.
I start with Nihon-siki as a base, as it is an accepted standard and has a philosophy of regularity and unambiguity that is compatible with my goals.
The first and most obvious issue to address is the distinction between stable and unstable consonants. There are only three types of instability found in the standard fifty-sound tables:
- Palatalization, as with し [ɕi] from さ [sa];
- Affrication, as with つ [tsu] from た [ta];
- Bilabialization, as with ふ [ɸu] from は [ha].
I wish to maintain the visual similarity of the letters to their parent consonant columns, but nevertheless indicate that a change has occurred, so I use only slightly modified forms of the letters.
For palatalization, I follow the Slavic convention of writing a caron on the letter:
- し becomes ‹ši›
- じ becomes ‹ži›
- ち becomes ‹ťi›
- ぢ becomes ‹ďi›
For affrication, I use a hook on the ascender of the letter:
- つ becomes ‹ƭu›
- づ becomes ‹ɗu›
For bilabialization, which is a kind of affrication, I use a hook on the right leg of the ‹h›:
- ふ becomes ‹ꜧu›
Though only found in the examples given in the standard fifty-sound tables, these consonant phonemes may be combined with any vowel to give combinations such as ‘ꜧairu’ for ファイル or ‘ťou’ for ちょう; since the changed forms are explicitly indicated, the unmodified letters are now available for use in words like パーティー (‘PÂTÎ’, ‘party’) without ambiguity.
I shall preserve the relationship between kana and romanization of the ‘long o’, and thus avoid ambiguity between words like 大 and 王, simply by writing them in Latin characters as they would be in kana:
- おお/オオ becomes ‹oo›/‹OO›
- おう/オウ becomes ‹ou›/‹OU›
- オー becomes ‹Ô›
Other vowels are treated similarly. Since we write the vowels separately, a lack of assimilation into a long vowel can be written explicitly with a diæresis on the second vowel, as is standard in English, thus preserving the 王 (‘ou’, ‘king’) / 追う (‘oü’, ‘chases’) distinction from speech.
In Tokyo Japanese, it is sufficient to indicate the position of the tone drop on a word to distinguish from near-homophones. This can be done with a single diacritic; for its evocative shape, I choose the grave accent. Other dialects have more complex rules for pitch accent; they are not covered here, and I suggest that writers consider tone-marking optional for dialects that require more than one mark per mora.
Though I am not aware of any minimal pairs, for consistency and as an aid to the pronunciation of novices, pitch accent on katakana words using the 長音 ‘long vowel’ mark (ー) should also be indicated. This is somewhat trickier due to the representation of the two morae as a single character. To this end I find I must introduce one more diacritic, the acute accent. Addition of the acute accent to a vowel with a circumflex indicates that the tone drop is on the second mora; addition of the grave accent marks the first mora.
Irregular Kana Usage
There are a very few cases in Japanese where the kana used does not correspond to the pronunciation of the word. The most usual examples are the topic particle は and the destination/direction particle へ. These are written according to the pronunciation, and then marked with a superscript letter indicating that the kana is written as if the superscript letter were present and replaced the consonant immediately before (if any). For example:
- は becomes ‹wʰa›
- を becomes ‹ʷo›
- へ becomes ‹ʰe›
- ゑ becomes ‹ʷe›
Most romanization systems, if they support the representation at all, are ambiguous when it comes to sounds that can be written either as a single kana or as a combination of a full vowel with a small kana. I suggest alleviating this by indicating the rarer full-vowel variants with a dot above:
- わ is written ‹wa›, while うぁ is written ‹ẇa›
- や is written ‹ya›, while いぁ is written ‹ẏa›
- ヷ is written ‹VA›, while ヴぁ is written ‹V̇A›
Some other romanization systems write ん either as ‹n› or as ‹n'› depending on whether it occurs in a position in which it could be confused with the consonant of a ‘n’-series kana, i.e. before a vowel or ‹y›. For simplicity's sake, I advise consistently writing n'.
For ease of input and as a fallback in technology-starved contexts, I attach an automatically-translatable ASCII encoding of the proposed system. This isn't recommended as the primary form of the system, as it is by necessity a little less regular, a little less language-neutral, and a little more cluttered, but it will serve in a pinch.
To convert the system to ASCII:
- replace every palatalization caron with a following letter ‹y› (ši → syi, ťou → tyou)
- replace every affrication hook with a following letter ‹s› (for unvoiced consonants) or ‹z› (for voiced consonants) (ƭu → tsu, ɗu → dzu)
- replace every ‹ꜧ› with the sequence ‹ph› (ꜧu → phu)
- replace every diæresis mark with a preceding hyphen (oü → o-u)
- replace every dot above with a following apostrophe (ẇa → w'a)
- replace every grave accent with a following backtick (sàke → sa`ke)
- replace every acute accent with a following apostrophe (BÎSYÎDYÎ́ → BÎSYÎDYÎ') (not ambiguous because no full-vowel combination will have an acute)
- replace every circumflex with a following caret (BÎSYÎDYÎ' → BI^SYI^DYI^')
- replace every superscript consonant with its non-superscript equivalent in parentheses (wʰa → w(h)a, ʷo → (w)o)
To convert the ASCII representation into the system, invert the conversion.
|あ a||か ka||が ga||さ sa||ざ za||た ta||だ da||な na||は ha||ぱ pa||ば ba||ま ma||や ya||ら ra||わ wa||ん n'|
|い i||き ki||ぎ gi||し ši||じ ži||ち ťi||ぢ ďi||に ni||ひ hi||ぴ pi||び bi||み mi||り ri||ゐ wi|
|う u||く ku||ぐ gu||す su||ず zu||つ ƭu||づ ɗu||ぬ nu||ふ ꜧu||ぷ pu||ぶ bu||む mu||ゆ yu||る ru|
|え e||け ke||げ ge||せ se||ぜ ze||て te||で de||ね ne||へ he||ぺ pe||べ be||め me||れ re||ゑ ʷe|
|お o||こ ko||ご go||そ so||ぞ zo||と to||ど do||の no||ほ ho||ぽ po||ぼ bo||も mo||よ yo||ろ ro||を ʷo|
|Ø||/a/ あ ‹a›||/i/ い ‹i›||/u/ う ‹u›||/e/ え ‹e›||/o/ お ‹o›||/ja/ や ‹ya›||/ju/ ゆ ‹yu›||/jo/ よ ‹yo›|
|k||/ka/ か ‹ka›||/ki/ き ‹ki›||/ku/ く ‹ku›||/ke/ け ‹ke›||/ko/ こ ‹ko›||/kja/ きゃ ‹kya›||/kju/ きゅ ‹kyu›||/kjo/ きょ ‹kyo›|
|g||/ga/ が ‹ga›||/gi/ ぎ ‹gi›||/gu/ ぐ ‹gu›||/ge/ げ ‹ge›||/go/ ご ‹go›||/gja/ ぎゃ ‹gya›||/gju/ ぎゅ ‹gyu›||/gjo/ ぎょ ‹gyo›|
|s||/sa/ さ ‹sa›||/si/ せぃ ‹si›||/su/ す ‹su›||/se/ せ ‹se›||/so/ そ ‹so›|
|š||/ɕa/ しゃ ‹ša›||/ɕi/ し ‹ši›||/ɕu/ しゅ ‹šu›||/ɕe/ しぇ ‹še›||/ɕo/ しょ ‹šo›|
|z||/za/ ざ ‹za›||/zi/ ぜぃ ‹zi›||/zu/ ず ‹zu›||/ze/ ぜ ‹ze›||/zo/ ぞ ‹zo›|
|ž||/ʑa/ じゃ ‹ža›||/ʑi/ じ ‹ži›||/ʑu/ じゅ ‹žu›||/ʑe/ じぇ ‹že›||/ʑo/ じょ ‹žo›|
|t||/ta/ た ‹ta›||/ti/ てぃ ‹ti›||/tu/ とぅ ‹tu›||/te/ て ‹te›||/to/ と ‹to›|
|ť||/tɕa/ ちゃ ‹ťa›||/tɕi/ ち ‹ťi›||/tɕu/ ちゅ ‹ťu›||/tɕe/ ちぇ ‹ťe›||/tɕo/ ちょ ‹ťo›|
|ƭ||/tsa/ つゃ ‹ƭa›||/tsi/ つぃ ‹ƭi›||/tsu/ つ ‹ƭu›||/tse/ つぇ ‹ƭe›||/tso/ つょ ‹ƭo›|
|ď||/dʑa/ ぢゃ ‹ďa›||/dʑi/ ぢ ‹ďi›||/dʑu/ ぢゅ ‹ďu›||/dʑe/ ぢぇ ‹ďe›||/dʑo/ ぢょ ‹ďo›|
|ɗ||/za/ づゃ ‹ɗa›||/zi/ づぃ ‹ɗi›||/zu/ づ ‹ɗu›||/ze/ づぇ ‹ɗe›||/zo/ づょ ‹ɗo›|
|n||/na/ な ‹na›||/ni/ に ‹ni›||/nu/ ぬ ‹nu›||/ne/ ね ‹ne›||/no/ の ‹no›||/nja/ にゃ ‹nya›||/nju/ にゅ ‹nyu›||/njo/ にょ ‹nyo›|
|h||/ha/ は ‹ha›||/hi/ ひ ‹hi›||/hu/ はぅ ‹hu›||/he/ へ ‹he›||/ho/ ほ ‹ho›||/hja/ ひゃ ‹hya›||/hju/ ひゅ ‹hyu›||/hjo/ ひょ ‹hyo›|
|ꜧ||/ɸa/ ふぁ ‹ꜧa›||/ɸi/ ふぃ ‹ꜧi›||/ɸu/ ふ ‹ꜧu›||/ɸe/ ふぇ ‹ꜧe›||/ɸo/ ふぉ ‹ꜧo›|
|p||/pa/ ぱ ‹pa›||/pi/ ぴ ‹pi›||/pu/ ぷ ‹pu›||/pe/ ぺ ‹pe›||/po/ ぽ ‹po›||/pja/ ぴゃ ‹pya›||/pju/ ぴゅ ‹pyu›||/pjo/ ぴょ ‹pyo›|
|b||/ba/ ば ‹ba›||/bi/ び ‹bi›||/bu/ ぶ ‹bu›||/be/ べ ‹be›||/bo/ ぼ ‹bo›||/bja/ びゃ ‹bya›||/bju/ びゅ ‹byu›||/bjo/ びょ ‹byo›|
|m||/ma/ ま ‹ma›||/mi/ み ‹mi›||/mu/ む ‹mu›||/me/ め ‹me›||/mo/ も ‹mo›||/mja/ みゃ ‹mya›||/mju/ みゅ ‹myu›||/mjo/ みょ ‹myo›|
|r||/ra/ ら ‹ra›||/ri/ り ‹ri›||/ru/ る ‹ru›||/re/ れ ‹re›||/ro/ ろ ‹ro›||/rja/ りゃ ‹rya›||/rju/ りゅ ‹ryu›||/rjo/ りょ ‹ryo›|
|w||/wa/ わ ‹wa›||/wi/ ゐ ‹wi›||/we/ ゑ ‹we›||/wo/ を ‹wo›|
|ẇ||/wa/ うぁ ‹ẇa›||/wi/ うぃ ‹ẇi›||/we/ うぇ ‹ẇe›||/wo/ うぉ ‹ẇo›|
|v||/va/ ヷ ‹va›|
|v̇||/va/ ゔぁ ‹v̇a›||/vi/ ゔぃ ‹v̇i›||/vu/ ゔ ‹v̇u›||/ve/ ゔぇ ‹v̇e›||/vo/ ゔぉ ‹v̇o›|
|ẏ||/ja/ いぁ ‹ẏa›||/ju/ いぅ ‹ẏu›||/je/ いぇ ‹ẏe›||/jo/ いぉ ‹ẏo›|
|n'||/ɴ̩/ ん ‹n'›|