tengwar
v1.2.6
Published
This is a Tengwar transcriber suitable for transcribing Sindarin Elvish from a phonetic encoding of the Latin alphabet, to the General Use mode of the Tengwar. It is written in JavaScript and is suitable for use as:
Downloads
392
Readme
This is a Tengwar transcriber suitable for transcribing Sindarin Elvish from a phonetic encoding of the Latin alphabet, to the General Use mode of the Tengwar. It is written in JavaScript and is suitable for use as:
- A plain script in a web page,
tengwar.min.js
. - A CommonJS module as used by Node or Mr, with the NPM package name
tengwar
.
Using the Script
The script searches the document for elements with the tengwar
class.
The class must also include either parmaite
or annatar
to select the
rendering font. This is not merely for the purpose of applying the
appropriate web font, but also instructs the script on which bindings to
use for kerning tehtar. The body of a tengwar
class must be rendered
with the included Tengar Annatar variant webfont or Tengwar Parmaitë
using the included tengwar-annatar.css
or tengwar-parmaite.css
.
If the element has a data-tengwar
property, that property is expected
to contain phonetic letters from the latin alphabet and gets transcribed
into bindings for the Tengwar Anntar font in the General Use mode,
popular for Sindarin and English. The script populates the element's
inner HTML with the font bindings, rendering the desired tengwar text
visible.
class="tengwar annatar"
If the element has a data-mode
property, the latin letters
are instead transcribed into key bindings through the
Classical mode, popular for Quenya, or the mode of Beleriand. Various
options can also be applied.
data-mode="general-use no-ach-laut reverse-curls"
data-mode="classical reverse-curls"
data-mode="beleriand"
data-mode="general-use black-speech"
If the element has a data-encoded
property, the value is expected to
be a description of the tengwar and tehtar to display like
romen:a;ungwe:a;romen:o;numen
for "Aragorn" in the General Use mode.
data-encoded="romen:a;ungwe:a;romen:o;numen"
Of course, a page can bypass the whole automated transcription process by statically populating the element with the desired key bindings and using neither of these data properties.
The script checks for modern browser features and stops if the necessary features are not present.
Using the Modules
tengwar/general-use
transcribes phonetic latin letters, as Tolkien wrote it, into Tengwar Notation in the General Use mode, suitable for Sindarin and many other languages.transcribe(text, options)
to key bindings for the font. Tengwar Annatar by default.encode(text, options)
to Tengwar Notationparse(text, options)
to Tengwar Object NotationmakeOptions(options)
font
defaults to the TengwarAnnatar module.block
whether to include HTML tags for paragraphs and line breaks.plain
whether to exclude all HTML from the output, making it suitable for plain text..blackSpeech
: In the Black Speech of the ring inscription, the "o" and "u" curls are reversed, medial "r" is ore before consonants in addition to final "r", and "sh" and "gh" used extended tengwar. This impliesreverseCurls
andmedialOre
.doubleNasalsWithTildeBelow
: Many tengwa can be doubled in General Use mode by placing a tilde above the tengwa, and many tengwa can be prefixed with the sound of the corresponding nasal by putting a tilde below the tengwa. Tengwar that represent nasal sounds have the special distinction that either rule might apply in order to double their value.false
: by default, a tilde above doubles a nasaltrue
: a tilde below doubles a nasal
reverseCurls
: In the Black Speech of the ring inscription, among other samples, the "o" and "u" tehtar are reversed.false
: by default, the "o" tehta curls forward, and "u" backward.true
: "o" curls backward, "u" forward.
swapDotSlash
false
: by default, "i" is a dot and "e" is a slash.true
: "i" is a slash, "e" is a dot.
noAchLaut
false
: by default, "ch" is transcribed as ach-laut, the "ch" as in "Bach". "cc" is transcribed as "ch" as in "chew".true
: "ch" is interpreted as the "ch" as in "chew".
sHook
false
: by default, "is" is silme-nuquerna with an I tehta.true
: "is" is a short carrier with an I tehta and S hook.
tengwar/classical
transcribes phonetic latin letters into Tengwar Notation in the Classical mode, most commonly used for Quenya.transcribe(text, options)
to key bindings for the font. Tengwar Annatar by default.encode(text, options)
to Tengwar Notationparse(text, options)
to Tengwar Object NotationmakeOptions(options)
font
defaults to the TengwarAnnatar module.block
whether to include HTML tags for paragraphs and line breaks.plain
whether to exclude all HTML from the output, making it suitable for plain text..viyla
: In the earlier forms of the mode, the tengwa "vilya" represented the sound of the letter V. The tengwa "vala" eventually replaced its role and "vilya" was renamed "wilya", and used for the sound of W, consonantal U.false
: by default "wilya" serves for W and "vala" for V.true
: "vilya" serves for V, and W is interpreted as the vowel U.
reverseCurls
: In the Black Speech of the ring inscription, among other samples, the "o" and "u" tehtar are reversed.false
: by default, the "o" tehta curls forward, and "u" backward.true
: "o" curls backward, "u" forward.
iuRising
: In the Third Age, IU is a rising diphthong, meaning that the stress is on the second sound. Whether to represent a rising diphthong in the same fashion as other diphthongs is a matter of conjecture.false
: by default, IU is rendered as the I tehta over "ure", the U tehta.true
: IU is rendered as the tengwa "anna" with a Y tehta below, and a U tehta above.
classical
: Before the Third Age (as defined by the Namarië) transcribers dealt with R and H differently. R can be rendered as either "romen" or "ore", but the rules differ. In the classical period, R is interpreted as "ore" only when it appears between vowel sounds. In the Third Age, R is interpreted as "ore" before consonants and at the end of words. The treatment of H is more complex and I have only given it a rough draft.false
: by default, we transcribe in the pattern of the Namarië poem, where "ore" is used finally and before consonants.- H is interpreted as "hyarmen".
- HY is interpreted as "hyarmen" with the underposed "y" tehta.
- HW and WH are interpreted as "hwesta".
- CH is interpreted as "harma".
- HT is interpreted as "harma" followed by "tinco". Therby, HT implies CHT.
- HL is interpreted as "halla" followed by "lambe".
- HR is interpreted as "halla" followed by "romen".
true
: "ore" appears only between vowels. The treatment of "H" depends on whether "harma" has been introduced yet.
harma
: In the Classical period, "hyarmen" implied the following-Y. Then "hyarmen" served as breath-H medially, and "harma" served as breath-H initially and was renamed "aha" in that role.false
: by default- H is interpreted as "halla" in all positions
- HY is interpreted as "hyarmen" with underposed "y".
- HT still implies CHT so treated as "harma" as above.
- CH, HL, HR, and HW (and WH) are not affected.
true
: the oldest form of the mode- H initial is interpreted as "harma"
- H medial is interpreted as "hyarmen"
- HY is interpreted as "hyarmen"
- HT still implies CHT so treated as "harma" as above.
- CH, HL, HR, and HW (and WH) are not affected.
tengwar/beleriand
: transcribes phonetic latin letters into Tengwar Notation in the mode of Beleriand, which is suitable for Sindarin and uses full tengwar for most vowels, instead of tehtar.transcribe(text, options)
to key bindings for the font. Tengwar Annatar by default.encode(text, options)
to Tengwar Notationparse(text, options)
to Tengwar Object NotationmakeOptions(options)
font
defaults to the TengwarAnnatar module.block
whether to include HTML tags for paragraphs and line breaks.plain
whether to exclude all HTML from the output, making it suitable for plain text..
tengwar/tengwar-annatar
: Translates Tengwar Object Notation into key bindings for Johan Winge’s Tengwar Annatar font. Provides themakeColumn
primitive which is aware of how a column of tengwar and tehtar can transform to accommodate additional tehtar with this font.transcribe(tengwarObjectNotation, options)
: to Tengwar Annatar key bindingsplain
: plain text, no markupblock
: block markup, with paragraph and line break tags
makeColumn(tengwa, above, below)
canAddAbove()
addAbove(tehta)
canAddBelow()
addBelow(below)
addFollowing(following)
addTildeAbove()
addTildeBelow()
addError(error)
notation
encode(tengwarObjectNotation)
: to Tengwar Notationdecode(tengwarNotation, makeColumn)
: to Tengwar Object Notation.decodeWord(tengwarNotation makeColumn)
: to Tengwar Object Notation for just one word (no nested arrays).
Tengwar Notation
Tengwar Notation is useful for succinctly representing the first stage of transcription, before translation to key bindings for a particular font. The notation uses the names of the tengwa followed by a list of tehtar in a consistent order:
- column =
- tengwa
- ":" if there are any following tehtar
- tehtar delimited by ","
- tehta above if applicable
- tehta below if applicable
- following tehta if applicable
- "tilde-above" if applicable
- "tilde-below" if applicable
- word = column delimited by ";"
- sentence = word delimited by " "
- stanza = sentence delimited by "\n"
- paragraph = stanza delimited by "\n\n"
- section = paragraph delimited by "\n\n\n+"
The notation is useful for manually describing a transcription, either to override the transcriber, or for testing a transcriber.
Tengwar Object Notation
Tengwar Object Notation represents a word of Tengwar as an array of objects. Each object has properties,
tengwa
the name of one of the tengwar or punctuation mark in my obtuse pidgin of punctuation names: "comma", "full-stop", "exclamation-point", "question-mark", "open-paren", "close-paren", "flourish-left", or "flourish-right". "vilya" is always represented as "wilya" and "aha" is always "harma", regardless of what name is appropriate for the mode.above
may be a tehta including "a", "e", "i", "o", "u", "ó", or "ú". Note that "á", "é", and "í" are not supported diacritics.below
may be "y".following
a tehta like "s", "s-inverse", "s-extended", or "s-flourish".tilde-above
boolean.tilde-below
boolean.
Words are wrapped in an array to make a sentence. Sentences are wrapped to make paragraphs. Paragraphs are wrapped to make sections. Somehow I’ve neglected stanzas within paragraphs. This will be remedied in a future version, and the nodes will probably be revised to be more sophisticated than merely nested arrays.
A font module must have a makeColumn
function that produces objects
with these properties and the attendant methods as described for the
Tengwar Annatar module above.