4. Script and Spelling in the Old Japanese Record
Old Japanese is recorded in Chinese characters. A graph can identify a Japanese word by its meaning, supply part of its pronunciation, or take part in a spelling that combines both methods. The choice depends on the passage, the writing tradition and sometimes the individual witness. This chapter distinguishes the written signs, the readings assigned to them and the normalized forms used in the grammar.
Chinese characters in Japanese passages
The language of a passage and the script in which it is written are separate properties. A passage in Chinese and a passage in Japanese can both consist of Chinese characters. Identify Japanese from evidence such as the order of words, particles and inflections recoverable in the passage, and the occurrence of continuous sound spellings. The shape of an individual graph gives little evidence about the language of a passage. A phonographically written Japanese name inside a Chinese sentence establishes the reading of the name while leaving the language of the surrounding sentence to be determined from its syntax.
The distribution of scripts and languages is easiest to see across different kinds of passage within one work. The Kojiki includes prose with varying degrees of reliance on Chinese writing conventions as well as songs in which character sequences give much more direct access to Japanese wording. The Nihon shoki includes Chinese annalistic prose, Japanese songs, and explanations or glosses that provide readings for particular items. Establish the scope of each reading before applying it to adjacent text. A Japanese translation or traditional Japanese reading of Chinese prose is evidence for how the prose was read, with a different evidential status from a Japanese wording spelled out in the text itself. (Chamberlain 1882, § Preface and the songs of the upper book; Aston 1896, § Introduction and books I–III)
A graph is the basic unit for describing this writing, but what it represents varies. A semantic graph can stand for a whole Japanese word, including several syllables. A phonographic graph commonly supplies one mora. In a mixed spelling, the division between graphs can cut across the division between a stem and its ending. Chinese character structure also operates at a different level. The components inside a single character remain parts of that character even if one of them has a phonetic function in the history of the Chinese graph.
| Unit under discussion | What can be counted or identified |
|---|---|
| Graphic sign | A character, repetition sign or other mark at a position in the witness |
| Spelling unit | One or more signs interpreted together, as when a graph has an associated reading gloss |
| Sound unit | A mora or a longer stretch of the Japanese reading, depending on how the graphs are used |
| Grammatical unit | A stem, affix, particle, word or larger expression assigned in the analysis |
A graph boundary therefore supplies a place to align a transcription with a reading, but it does not necessarily supply a word boundary. A single word can require several graphs, and a particle can have a spelling of its own even where it forms a close grammatical and prosodic unit with the preceding expression. Spaces between words in romanization are editorial. Spaces or gaps visible in a witness can have layout significance and should be described on the evidence of that witness.
The eighth-century use of Chinese characters antedates the independent hiragana and katakana orthographies. Later kana are useful in editions as glosses and reading aids. Their presence on an edited page or in a later copy also raises a question about which layer they belong to. A kana reading printed next to an old graph supplies an editor's or annotator's reading unless the edition identifies it as part of the material being transcribed. The same applies to punctuation, accent marks and other additions.
For a simple illustration, 山 can stand for the Japanese noun yama. The graph supplies a lexical identification in a Japanese context, while the identification of the passage as Japanese comes from the larger text.
| Form | Romanization | Gloss | Translation |
|---|---|---|---|
| 山 | yama | mountain | ‘mountain’ (constructed) |
Semantic, phonographic and mixed uses
A semantic use selects a graph through a relation between its meaning and the Japanese expression to be read. The Japanese reading supplies the sound form. Context may be needed to choose among Japanese words for which the same graph would be semantically appropriate, or among inflected forms of one word. In a prose passage that represents content densely, an edition can also supply particles or endings required by its reading. Those supplied elements have weaker graphic support than elements represented by their own signs.
A phonographic use selects a graph for its sound value in the spelling tradition. Its ordinary meaning can be irrelevant to the Japanese expression. In a continuous phonographic spelling, the graphs can follow the sequence of morae closely enough to show a particle or ending. This makes such spellings especially useful for distinctions that a semantic graph leaves unexpressed. A phonographic spelling still requires a choice of graph values, a division into linguistic units and an interpretation of the resulting words.
A mixed spelling has semantic and phonographic material in the same expression or passage. One common arrangement uses a semantic graph for the lexical item and a phonogram for a following particle. A verb can likewise combine a lexical graph and a phonographically expressed terminal portion. Another arrangement puts sound glosses alongside semantically written material. That layout records two kinds of evidence for the same expression, and the gloss must be aligned with the expression it actually explains. The alternation among these methods is visible across narrative, song and glosses in the early chronicles. (Chamberlain 1882, § Preface and the songs of the upper book; Aston 1896, § Introduction and book I)
| Form | Romanization | Gloss | Translation | Use in the illustration |
|---|---|---|---|---|
| 山 | yama | mountain | ‘mountain’ (constructed) | Semantic |
| 夜麻 | yama | mountain | ‘mountain’ (constructed) | Phonographic throughout |
| 山尓 | yama ni | mountain LOC | ‘at the mountain’ (constructed) | Semantic noun and phonographic particle |
| 夜麻尓 | yama ni | mountain LOC | ‘at the mountain’ (constructed) | Phonographic noun and particle |
In 夜麻 the assigned values are ya and ma. The gloss ‘mountain’ applies to their combination. In 山尓, the first graph has the whole reading yama and the second supplies ni. Both rows that contain the particle have the same grammatical segmentation, even though the graphic boundary after the noun falls at a different point in the character sequence. LOC is the locative gloss for ni in this example.
The spelling of a verb illustrates another alignment. A phonogram for a final syllable can cover the final consonant of a stem and the following inflectional vowel. In the constructed spelling below, the final graph supplies tu, while the morphological analysis places the boundary between t and u. Assigning a full syllable to the phonogram is compatible with glossing a vowel as the inflection.
| Form | Romanization | Gloss | Translation |
|---|---|---|---|
| 立都 | tat-u | stand-FIN | ‘stands’ (constructed) |
Here FIN labels a finite ending. The example gives a possible mixed alignment. The same graph 都 could receive a semantic analysis in a context that supports a word for a capital or city. Its position after the lexical verb graph is part of the evidence for its phonographic function here.
For an unfamiliar sequence, start with values established for the witness or text group. Check whether a proposed reading gives a plausible Japanese sequence and accounts for the signs around it, including any spellings of particles and endings. Use repeated occurrences, parallel passages and glosses to test that reading. A character's dictionary meanings can suggest a semantic reading, and known sound values can suggest a phonographic one. An isolated match of either sort leaves room for more than one analysis.
Some spellings invite the reader to recover a word through the significance or arrangement of the signs. A rebus may use a graph for the sound of a Japanese word associated with it. More elaborate play can require an intermediate interpretation before the sounds become clear. Classify these by the operation actually needed to recover the reading. A phonogram chosen for a native sound value can be routine in one tradition even if an inventive use of the same principle would count as wordplay elsewhere. Commentary on a poem can help establish the intended reading, while the graphs themselves remain the evidence for its written form. (Keichū 1690, § commentary on Man’yōshū book I)
Sound-based and native-reading phonograms
There are two principal routes from a Chinese character to a Japanese phonographic value. An on phonogram uses a value from a Chinese-derived reading. A kun phonogram uses the sound of a Japanese word associated with the graph by meaning. Both are phonograms when the resulting sound is used to write another expression. The terms identify the source of the value. They do not classify every occurrence of the character.
| Source of value | How the spelling works | Evidence that helps establish the value |
|---|---|---|
| Chinese-derived reading, on | The spelling convention assigns a sound value based on a reading of the character borrowed from Chinese | Uses of the graph in clear Japanese sound spellings and the relevant historical reading tradition |
| Native reading, kun | The spelling convention assigns the sound of a Japanese word associated with the graph, or a conventional part of that sound | The native reading and repeated spellings in which that reading contributes sound to another expression |
For an on phonogram, a modern Sino-Japanese reading can be a clue to the history of the graph, but the spelling value must be established for Old Japanese. The process involves a historical Chinese reading tradition and Japanese adaptation. A phonogram represents the value as used by Japanese writers. Features of a source reading that do not correspond to the assigned Old Japanese value do not carry over into the manual's romanization.
For a kun phonogram, the writer starts from a meaning association. The sound then works independently of that meaning. A native word of one mora can supply a value of one mora. A longer native reading can supply a longer sequence, or part of it can yield a shorter value if the usage supports that analysis. Record which part of the reading is conventionally available, since the possibility of cutting a longer reading at a particular point is insufficient evidence that a scribe used the resulting value.
| Form | Romanization | Gloss | Translation | Reading illustrated |
|---|---|---|---|---|
| 麻 | ma | MA | ‘the sound ma’ (constructed) | Chinese-derived phonographic value |
| 手 | te | hand | ‘hand’ (constructed) | Semantic use, with a Japanese lexical reading |
| 手 | te | TE | ‘the sound te’ (constructed) | Possible native-reading phonographic use in a sound spelling |
Capitalized MA and TE describe the function of isolated graphs in the table. In a word or sentence gloss, gloss the expression written by the phonogram sequence. A native-reading value and a semantic use can give the same romanization, as with 手. To determine the function, ask whether the Japanese expression at that position contains a word meaning ‘hand’ or merely the sequence te. In the latter case, a position inside a larger word whose meaning has no connection with hands supports the phonographic reading.
Several consequences follow when reading an unfamiliar spelling.
- A graph can have semantic uses and several phonographic values. Select among them in context, using the readings established for that text.
- A phonographic value can be written by several graphs. The relevant correspondences belong to the spelling tradition and may be distributed differently in another witness.
- A long Japanese reading can belong to a single graph. In the other direction, multiple graphs can be needed to write a single grammatical word.
- The same spelling can leave a lexical distinction unresolved. Two words with the same supported pronunciation require independent evidence such as syntax, meaning or a parallel passage.
A graph chosen for its sound can also have associations that matter to the interpretation of a name or poem. A plausible association alone does not determine the spelling function. First establish the reading that the graph contributes. Then evaluate any further semantic effect in that context. This preserves the difference between a reading recoverable from the phonograms and a literary interpretation of their selection.
Man’yōgana inventories and spelling variation by text
Man’yōgana is the conventional name for the use of Chinese characters as Japanese phonograms in early texts. The name refers to the Man’yōshū, where the practice is conspicuous, and applies to such uses in other texts as well. It describes a function of characters. A text conventionally described as using man’yōgana can also have semantic spellings, and a graph counted as man’yōgana in one position can be used semantically elsewhere. The commentary tradition of the Man’yōshū and the songs and glosses transmitted in the chronicles provide different kinds of evidence for the written readings. (Keichū 1690, § commentary on Man’yōshū book I; Chamberlain 1882, § songs of the upper book; Aston 1896, § book I)
There is no single table of character-to-sound correspondences that can be applied without qualification to every Old Japanese text. An inventory for a text records the characters used phonographically there, the values supported in those uses and any distributional limits. Where some occurrences come from a gloss or a distinct type of passage, record that scope. An inventory based on one edition can also partly reflect the editor's choice of readings or normalization of graphs. It needs to identify the underlying text or witness if it is to support claims about scribal practice.
The examples here show individual correspondences. They leave the witness-specific inventories open.
| Form | Romanization | Gloss | Translation | Spelling point |
|---|---|---|---|---|
| 加 | ka | KA | ‘the sound ka’ (constructed) | One phonographic spelling of ka |
| 可 | ka | KA | ‘the sound ka’ (constructed) | Another phonographic spelling of ka |
| 夜 | ya | YA | ‘the sound ya’ (constructed) | A sound value used in the spelling of a longer word |
| 尓 | ni | NI | ‘the sound ni’ (constructed) | A sound value that can express the locative particle |
The presence of two signs for ka leaves open who used each and where. Some apparent alternations are choices among distinct characters for the same value. Others are variant shapes of one character, which belong in a description of the witness's handwriting. A third kind of alternation uses distinct characters because the spelling system distinguishes sound classes later written alike. Those signs are interchangeable only if the relevant text actually treats the classes alike.
Comparisons across texts should align like units. A largely phonographic song can show the reading of a particle or a vowel class where a semantically written narrative expression has no corresponding phonogram. A text with many semantic spellings may therefore have fewer readable instances of a spelling distinction. The silence of such a passage leaves the class unobserved. Across witnesses to the same text, a substitution can reflect a copyist's choice, a different reading or an editor's normalization. To make the substitution evidence of any one of these, establish which witness shows it and where in its transmission it occurs.
The comparison can be made at several scales. A song transmitted inside a chronicle may use graph values differently from a gloss elsewhere in the same work. Collections can preserve material from different areas or different periods. When a pattern is local to a group of poems, describe that group before extending a statement to the collection. In particular, apply claims about the full set of paired spellings only to the traditions in which the pairs are established. Closely related texts can give different evidence for an individual pair. (Hashimoto 1917, § discussion of Ishizuka's character groupings)
Several editorial presentations are possible for a spelling such as 夜麻. A character transcription gives the two graphs, a transliteration of their assigned values gives ya-ma, and a linguistic reading gives yama. The division in ya-ma here represents the mapping between graphs and morae. The division does not assert that the word is composed of two morphemes. A semantic spelling 山 gives the same reading with only one graph to align. When the distinction matters, keep the graphic form alongside the reading.
The surviving texts show spelling conventions established through repeated practice. There was no uniform kana orthography governing these eighth-century spellings. Modern kana spelling rules and later historical kana conventions belong to editions or later reading traditions when used to present the early texts. The manual applies its own consistent conventions to romanized forms while retaining significant graphic evidence in the transcription.
The kō and otsu spelling distinctions
In certain syllables, phonographic spellings fall into two groups even though later Japanese writes the syllables alike. The traditional labels are kō 甲 and otsu 乙. This manual uses subscript ₁ for kō and subscript ₂ for otsu. The subscript applies to the whole syllable immediately preceding it. The subscripts are part of the romanization. The distinction in the original spelling is conveyed by the selection of characters. The analysis rests on the distribution of their uses, including regular correspondences between particular lexical or grammatical forms and particular sets of phonograms. (Hashimoto 1917, § discussion of Ishizuka's character groupings)
The following table gives the full set of paired syllables used in the manual's account of the Western Old Japanese distinctions, counting mo where the pair is preserved. The consonant letters are those of the manual's romanization, so the rows with p correspond to sounds whose later standard Japanese reflexes are in the h series. A dash indicates that the table contains no paired syllable for that consonant and vowel combination. It says nothing about whether an unpaired syllable occurs there.
| Initial | Paired i syllables | Paired e syllables | Paired o syllables |
|---|---|---|---|
| k | ki₁, ki₂ | ke₁, ke₂ | ko₁, ko₂ |
| g | gi₁, gi₂ | ge₁, ge₂ | go₁, go₂ |
| p | pi₁, pi₂ | pe₁, pe₂ | — |
| b | bi₁, bi₂ | be₁, be₂ | — |
| m | mi₁, mi₂ | me₁, me₂ | mo₁, mo₂ |
| s | — | — | so₁, so₂ |
| z | — | — | zo₁, zo₂ |
| t | — | — | to₁, to₂ |
| d | — | — | do₁, do₂ |
| n | — | — | no₁, no₂ |
| y | — | — | yo₁, yo₂ |
| r | — | — | ro₁, ro₂ |
The distinction in mo is recoverable in the Kojiki. It is generally unavailable in the later textual spelling traditions. The table gives the distinctions shown across the Western material, with this restriction on mo. Apply the inventory to a particular passage only to the extent that its region and spelling tradition support it. Other distinctions in the transcription, such as initial e and ye where separately identified, are written with their own letters. The kō and otsu labels apply only to the pairs in the table. (Hashimoto 1917, § discussion of the distribution of the paired spellings)
| Form | Romanization | Gloss | Translation |
|---|---|---|---|
| 支 | ki₁ | KI₁ | ‘the sound ki of class 1’ (constructed) |
| 紀 | ki₂ | KI₂ | ‘the sound ki of class 2’ (constructed) |
| 古 | ko₁ | KO₁ | ‘the sound ko of class 1’ (constructed) |
| 許 | ko₂ | KO₂ | ‘the sound ko of class 2’ (constructed) |
These are illustrations of phonographic assignments. In an actual passage, first establish that the graph is used phonographically. A semantic use of 古 need not write ko₁. In a reading of a poetic expression, evidence from a recurring lexical item can help choose between spellings with similar-looking graphs. The sound-class interpretation also depends on an identification of the graph that is defensible independently of the expected word.
Record the class from the spelling when the graph identifies it. A semantic spelling can instead be read with a class established by phonographic evidence for that lexical or grammatical form elsewhere. In that case the normalized reading records a lexical identification. A diplomatic or graph-by-graph account still shows that the passage itself gives no class-bearing phonogram at that position. If a class is unresolved in the reading, use an unsubscripted syllable and state the uncertainty when it matters. An unsubscripted vowel also occurs in syllables outside the scope of the pairs, where no class assignment is called for.
The spelling classes have consequences for the analysis of inflection and word formation. Forms that agree in modern spelling can preserve different classes, and the class in an ending or at a stem boundary can help identify the form. A classification of that kind requires sound spelling evidence. Where a text only gives a semantic graph, several inflections may be consistent with the writing until a particle, construction or parallel spelling resolves the reading.
The subscripts allow the distinctions to be recorded without committing to particular phonetic values. A historical account can ask whether the spellings reflect distinct vowels, sequences, or contrasts with different historical origins in different syllables. A romanized form carrying a subscript remains usable whatever reconstruction is being assessed. The same convention allows spellings from editions with different phonological notation to be compared by class.
Small graphs and other conventions of senmyō writing
Senmyō are imperial proclamations. The writing style associated with their Japanese wording is commonly called senmyōgaki. It uses graphic size to help organize a mixed spelling. Full-size characters ordinarily convey lexical material, frequently by semantic writing. Smaller phonograms express material such as particles, inflectional portions of words and other grammatical elements. The reader follows the order of the signs and integrates both sizes into a continuous Japanese reading. In related ritual writing, forms transmitted with grammatical material spelled out are likewise important evidence for reading connected Japanese. Ritual texts have their own transmission and editorial conventions. (Satow 1879, § introductory account of the ritual texts)
The distinction can be illustrated with a sequence whose characters are all available at the same size in plain text. Suppose 山 is full-size and 尓 small. The size information associates the full-size graph with the noun and the small graph with its particle.
| Form | Graphic sizes in the illustration | Romanization | Gloss | Translation |
|---|---|---|---|---|
| 山尓 | Full-size 山, small 尓 | yama ni | mountain LOC | ‘at the mountain’ (constructed) |
| 立都 | Full-size 立, small 都 | tat-u | stand-FIN | ‘stands’ (constructed) |
In the second row the small graph supplies the terminal syllable tu. It can signal inflectional material while writing a syllable that includes the final consonant of the stem. Thus size helps a reader interpret the spelling even where it cannot give an exact morphological division. Several small graphs in succession can spell more extensive grammatical material, and an unexpressed element still requires the support of the reading rather than a size assignment.
Size has to be interpreted together with position and the conventions of the witness. Reduced graphs written in the main sequence can be part of the proclaimed wording. Reduced graphs in an accompanying gloss can give a pronunciation or explanatory annotation. If the edition distinguishes these by arrangement, carry that distinction into the transcription. A string of Unicode characters with no size or placement information cannot recover the difference on its own.
Related uses of large and small writing may occur in a transmitted text or a print edition. When a printed edition changes placement or uses a typographic device to replace size, describe the device used in that edition. A comparison of two transcriptions should establish whether both preserve the original relative sizes, whether one substitutes an annotation and whether either changes the linear order. Variations can then be attributed to layout or editing where appropriate.
The spelling convention allows a lexical expression and its grammatical continuation to be differently visible. A small particle still belongs in the grammatical reading. A large semantic graph can leave its pronunciation partly dependent on context. Read the construction using both kinds of information, allowing for local departures from the ordinary pattern. Where a statement about the proclamation depends on a small graph, cite or reproduce the size information instead of inferring it from the linguistic interpretation alone.
Graphic variants, damaged signs and Unicode transcription
A witness can show different shapes for a character. Some differ only in handwriting or calligraphic treatment. Others are conventional variants, and others still may be distinct characters that an editor has brought under the same modern form. The level of transcription determines how much of this variation stays visible. A reading edition may use a standard graph identity. A diplomatic account records the distinction whenever it matters to identifying what the witness shows or to interpreting a spelling.
Distinguish similarity of shape from equality of sound value. Two visibly different characters can be phonograms for the same syllable. Two highly similar forms can be different characters with different values. When the result changes a reading, give the graph identification and the reading separately. A vowel class, lexical meaning or expected grammatical pattern can support a choice among graph identifications, alongside the physical traces and the scribe's practice.
| Condition in the witness or edition | Transcription treatment |
|---|---|
| An identifiable graph is written with ordinary variation in stroke shape | Use its established character identity in a character transcription. Preserve finer detail in a facsimile or description when relevant. |
| A recognizable variant matters to the spelling or to comparison of witnesses | Retain the variant if the transcription can represent it. Otherwise describe it with the graph identification. |
| A graph is recognizable but part of it is damaged | Transcribe the identification with the stated editorial qualification when the damage affects certainty. |
| Two or more readings of the traces remain possible | Keep the alternatives tied to the position. Give a preferred reading only when there is evidence for the preference. |
| A graph is illegible and no restoration is adopted | Mark the illegible position according to the declared notation. Give its extent only to the precision the witness supports. |
| The text is missing and an editor restores a graph or sequence | Explicitly mark the restoration. Identify a parallel witness or other basis when available. |
Damage to a visible graph and the physical loss of text are different observations. A date, formula or parallel passage may make a restoration probable without making any traces visible. Conversely, enough of a graph may survive to identify it even if the resulting passage remains difficult to read. In an account of the evidence, preserve both kinds of information. Where a restoration supplies an entire syllable or morpheme, its status carries through from the character transcription to the romanized reading and gloss.
Unicode represents character identities and provides separate symbols for some distinctions. It has limits as a description of a particular mark on a page. Unified Han characters can display with different shapes depending on the font or language settings. Some variants have distinct code points, and standardized variation sequences can identify some glyph variants. Both require support in the software and font used to display the transcription. An image or a description remains appropriate when the distinction is palaeographic, the needed form has no usable encoding, or reliable display cannot be assumed.
The same character code can be used for full-size and small senmyō graphs. Use layout, markup or a size annotation to preserve their distinction. Specialized marks may also have more than one aspect worth recording, such as their relative position and the scope they have in a reading. A plain character string can carry a text for searching while an accompanying diplomatic presentation preserves those aspects. A regularized search string cannot establish which variant a witness shows.
Repeated characters and repetition signs require separate treatment. Transcribe a visible repeat sign as such at the diplomatic level. In a reading, expand it to the unit it repeats if the scope and interpretation are clear. When it is unclear whether the sign repeats a graph, a reading or a larger sequence, preserve the ambiguity. Likewise, expand a scribal abbreviation only when the convention and its application in that position are established. A traditional reading that supplies material omitted in semantic writing is a reading of the construction. An explicit abbreviation in the witness provides a distinct kind of evidence for its expansion.
Editorial marks are easiest to interpret when each edition defines its own conventions. Brackets, a blank square and dots have been used with different meanings. One display may use a square for a known graph position whose character is unreadable, while another may represent a gap of uncertain extent. State the convention before using it, and reserve sigla for conditions that are recoverable from the presentation. Do the same for later additions, deletions and corrections. When two hands or two textual states can be identified, a running reading can adopt one while its transcription identifies both.
Diplomatic transcription, romanization and normalized readings
A diplomatic transcription, a graph-aligned transliteration and a normalized reading answer different questions. The diplomatic transcription records what is visible or recoverable in the written witness under a declared policy for variants and layout. A graph-aligned transliteration gives the assigned values in order and makes their alignment visible. A normalized reading gives the Japanese forms analyzed in the grammar. Some stretches support all three. In other stretches, a graph can be transcribed even when its contribution to the reading is uncertain.
| Representation | What it records | Treatment of editorial interpretation |
|---|---|---|
| Diplomatic transcription | Identified signs in order, with relevant variants, size, placement, damage and correction | Marks uncertain signs and supplied material according to a declared notation |
| Graph-aligned transliteration | Values assigned to signs or groups of signs in the represented passage | Associates each value with its written support and marks unresolved assignments |
| Normalized romanization | The Japanese reading, with linguistic segmentation and relevant sound distinctions | Supplies readings of semantic graphs and indicates significant uncertainty or restoration |
| Gloss and translation | The analysis of the Japanese reading and its sense in context | Glosses the words and morphemes in the reading and carries through uncertainty that affects the interpretation |
The manual gives original-script forms where they are relevant, followed by a Latin-script reading, a gloss and a translation. The romanization uses a, i, u, e, o as vowel letters, with the subscript convention for the paired spellings. It uses p for the historical consonant where the Old Japanese form is read with p, as in the paired pi and pe spellings. It writes consonant and vowel sequences such as si, ti and tu accordingly, without using later pronunciation conventions such as shi, chi and tsu for them. It preserves distinctions such as wi, we, wo and ye where the reading supports them. Letters for voiced consonants represent an established reading or analysis. They do not imply the presence of voicing marks on the original signs.
Combinations of Latin letters in the reading follow the segmentation of the Japanese form. In a sequence such as ye, y writes the initial consonant and e writes the vowel of that syllable. The Old Japanese text has no corresponding Latin digraph. A phonographic graph can stand for the entire syllable. A Latin spelling in a personal name, work title or source citation can follow the conventions for that name or title. In Man’yōshū and senmyō, for example, the macron belongs to the conventional romanization of the term. The subscripts ₁ and ₂ used for Old Japanese paired spellings carry information of a different kind.
| Form | Graph-aligned reading | Normalized romanization | Gloss | Translation |
|---|---|---|---|---|
| 夜麻尓 | ya · ma · ni | yama ni | mountain LOC | ‘at the mountain’ (constructed) |
| 山尓 | yama · ni | yama ni | mountain LOC | ‘at the mountain’ (constructed) |
| 立都 | ta · tu | tat-u | stand-FIN | ‘stands’ (constructed) |
The middle dot in the graph-aligned column separates values assigned to neighboring graphs. The space in yama ni separates the noun and particle in the grammatical presentation, without expressing a claim about spacing in the witness or about the particle's prosodic independence. In tat-u the hyphen marks an analyzed morpheme boundary and is reflected in the gloss stand-FIN. The graph-aligned reading of 立 identifies the portion contributed by the lexical graph in that spelling. The stem tat- is recovered from the whole sequence.
This treatment also distinguishes orthographic and morphophonemic information. A semantic graph used across several inflected forms can keep the lexical item recognizable while leaving some or all of its inflection to context. In a mixed spelling, a final phonogram can give a surface syllable that straddles the stem boundary. The morphological analysis may therefore recover the same stem from forms whose written terminal portions differ. Put the stem and ending in the normalized analysis, using the written forms to support the reading at each occurrence.
Sound changes and alternations at word boundaries present a related problem. Where a phonographic spelling records the form found in context, retain that form in the reading of the occurrence. The citation form of a lexical item can be given separately when it clarifies the analysis. Where the spelling is semantic and leaves the alternation unwritten, a phonological form supplied by the editor rests on the grammatical analysis or on parallel sound spellings. Spellings can also preserve different evidence across witnesses. A reading normalized for comparison should remain traceable to those spellings.
The choice of reading follows the purpose of the example. For a point about graph selection, retain the graph sequence and show the part of the reading it supports. For a point about kō and otsu, retain the subscripts and identify classes supplied from comparative evidence if the local spelling is semantic. For a point about syntax, give the segmented reading and gloss, marking any supplied particle or ending on which the analysis depends. A traditional reading can provide a coherent Japanese sentence from dense written material. Cite it as a reading where the exact wording could also be recovered differently. The chronicles' translations and the ritual and poetic commentary traditions illustrate why the distinction between a transmitted passage and its reading matters. (Chamberlain 1882, § Preface; Aston 1896, § Introduction; Satow 1879, § introductory account of the ritual texts; Keichū 1690, § commentary on Man’yōshū book I)
Romanization does not preserve a unique route back to the characters. In the table, yama ni corresponds to two sequences, and a third sequence could use another graph for an established sound value. Graph-to-reading conversion also depends on the lexical, grammatical and textual evidence. Retain the original-script form with the normalized reading whenever a conclusion depends on the choice of characters, the presence of a sound spelling or the distinction between an explicit graph and editorially supplied material.
How Mova handles script identification and spelling variants
tools/langseg assigns language labels to sentences and spans at level 1. Its Latin-script identification covers English and French using a Latin table, a lexicon for each language and a character 4-gram model. It also handles Hebrew-script varieties and scholarly transliteration of Hebrew or Aramaic. Chinese-character text falls under other. The language identification it provides does not establish which parts of such a text are Old Japanese. Romanizing a passage does not give tools/langseg an Old Japanese identification. Adding a further Latin-script language uses a lexicon and a Lang code, with langseg latin-lexicon available for lexicons from open Universal Dependencies training splits.
At level 2, tools/mova-parse uses a model for the language of each span when one exists. No level-2 model exists for ojp. The missing model is reported in the # parse comment. Affected tokens receive _ in LEMMA, UPOS, XPOS, FEATS, HEAD and DEPREL, with Parse=None in MISC. Mova has no Old Japanese tokenization, tagging, lemmatization or parsing and holds no annotated Old Japanese treebank. There is no measured baseline for ojp.
The package described in What every language gets asks for the same treebank baseline for each language. In the en crate, en train trains a tree parser and tagger on the language's Universal Dependencies training split after en xpos-fill. en parse-eval measures unlabeled and labeled attachment scores on the development split. Mova's frame, pragmatics and world components, prag and world, read English trees and have no Old Japanese parse to read.
Squid reads the manual as grammatical claims and queues claims that can be formalized. The distinctions in this chapter, including semantic and phonographic uses, text-specific sound values, paired spellings, small graphs and editorial readings, have no implemented rules in Mova. A queued claim awaits implementation. The first points needed for implementation are those that support a tokenizer, a tagger and a Universal Dependencies treebank baseline.
Sources for this chapter
- Chamberlain, Basil Hall. 1882. A Translation of the “Ko-ji-ki,” or “Records of Ancient Matters.” Transactions of the Asiatic Society of Japan 10, supplement. — public domain; not held in the library, to verify
- Aston, William George. 1896. Nihongi: Chronicles of Japan from the Earliest Times to A.D. 697. 2 vols. Supplement 1 to Transactions and Proceedings of the Japan Society, London. London: Kegan Paul, Trench, Trübner & Co. — public domain; not held in the library, to verify
- Hashimoto, Shinkichi. 1917. 国語仮名遣研究史上の一発見―石塚龍麿の仮名遣奥山路について [A discovery in the history of the study of Japanese kana usage: On Ishizuka Tatsumaro’s Kanazukai oku no yamamichi]. 帝国文学 [Teikoku Bungaku] 23(11). — public domain; not held in the library, to verify
- Keichū. 1690. 萬葉代匠記 [Man’yō daishōki, commentary on the Man’yōshū]. First manuscript recension, shokō-bon. — public domain; not held in the library, to verify
- Satow, Ernest Mason. 1879. Ancient Japanese Rituals. Part I. Transactions of the Asiatic Society of Japan 7. — public domain; not held in the library, to verify
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/ja/ojp/04-script-and-spelling/