A Cross-Cultural Romanization Scheme
Talk uses the Latin script with diacritics to encode most of Earth's natural language features, enough so that you can write every language using the same Latin-oriented system and be close enough to a realistic pronunciation, including nasalized vowels, tense consonants, clicks, and tones, amongst other things.
There are two forms:
- ASCII: For writing in a text editor without fancy symbols.
- Simplified: For reading with condensed characters and diacritics.
The goal of the simplified version is to make it as easy as possible to read text given basic knowledge of today's standard English writing practices. Of course you need to learn the basics if you really want to take advantage of it, but without learning anything, you still can get the gist of it.
| ascii | simplified |
|---|---|
| txando^ | txandȯ |
| surdjyo^ | suṙdjyȯ |
| HEth~Ah | ḥẹtɦạh |
| siqk | siṅk |
| txya@+a-a++u | txyá̖àa̋u |
| hwpo$kUi^mUno$s | hwpo̤kụïmụno̤s |
| sinho^rEsi | sinhȯṙẹsi |
| batO_'aH | batọ̄qaḥ |
| aiyuQaK | aiyuq̇aḳ |
| s'oQya&te | s'oq̇ya̰te |
| t!arEba | t̖aṙẹba |
| txhaK!EnEba | txhaḳ̖ẹnẹba |
| txh~im | txɦim |
| txy~h~im | txẏɦim |
| mh!im | mħim |
IPA and XSampa are useful because they are widespread amongst the linguistic community, and much has been encoded using (at least) the IPA encoding.
However, IPA is not that easy to understand or write for the average English speaker, and XSampa is too foreign for the same group of people (even though it's easy to write on a keyboard).
Our goal with Talk is to make an easy to write and easy to understand 3rd encoding which requires less expert knowledge. And a readable romanization form ("simplified" form) which an average English reader/speaker can get close to intuiting without more than a small handful of key notes, not requiring tons of special knowledge.
Also, Talk is not meant to replace the level-of-detail ("exactness") of IPA, it's goal is to only get like ~95% accuracy. This is because we believe real-world pronunciations a highly varied within even a single word of a language, and so being as exact as IPA is kind of misleading. If Talk is not exact enough for a specific rendering, you can use IPA, but otherwise Talk will be much easier to understand and use on both sides.
We are aware of competing standards haha. But for the websites we are planning on making, thinking of showing the general public IPA has (in our experiments / customer testing) so far revealed that it is either ignored or misunderstood, so we are making something that we think will be used and not misunderstood.
Almost every language learning resource on the web has their own version of writing the pronunciations of words, often like this:
pruh-nuhn-see-EY-shuhn
They use that because it's easier for people without linguistics knowledge to understand. So we sort of normalized things against a simplified system like this.
Here are the modifiers on consonants, vowels, and symbols (like punctuation), and how they look in ASCII and simplified form.
There are a few affixes on consonants and vowels. There are 5 tone affixes (extra low, low, neutral, high, and extra high), which can be combined in standard ways.
| category | symbol | meaning |
|---|---|---|
| consonant | h~ | Aspiration (when added after a consonant) |
| consonant | w~ | Labialization (when added after a consonant) |
| consonant | y~ | Palatalization (when added after a consonant) |
| consonant | G~ | Velarization (when added after a consonant) |
| consonant | Q~ | Pharyngealization (when added after a consonant) |
| consonant | ! | Makes consonant ejective |
| consonant | ? | Makes consonant implosive |
| consonant | @ | Makes consonant tense (korean) |
| consonant | . | Makes consonant a stop consonant (korean, so when you end on t., it is making the mouth shape of t but not really pronouncing it) |
| consonant | * | Finds the closes "click consonant" mapping (there are 5, d, p, k, l, t) |
| consonant | capital | Consonant variant |
| vowel | ^ | Stressed vowel |
| vowel | _ | Long vowel |
| vowel | ! | Short vowel |
| vowel | & | Nasal vowel |
| vowel | @ | Non-syllablic vowel |
| vowel | capital | Vowel variant |
| vowel | $ | Vowel variant |
| vowel | + | High tone (mandarin high) |
| vowel | ++ | Extra high tone |
| vowel | - | Low tone (mandarin low) |
| vowel | -- | Extra low tone |
| vowel | / | Rising tone (vietnamese sắc, mandarin rising) |
| vowel | // | Rising tone 2 (vietnamese ngã) |
| vowel | \ | Falling tone (vietnamese huyền, mandarin falling) |
| vowel | \\ | Falling tone 2 (vietnamese nặng) |
| vowel | /\ | Rising falling tone |
| vowel | \/ | Falling rising tone (vietnamese hỏi) |
| symbol | = | When preceding, can write a literal symbol, like =. is a period, =+ is a plus, etc.. |
Note: GitHub markdown doesn't really render the diacritics that nicely, some are misaligned. We will have a font to remedy this for websites.
| IPA | ascii | simplified |
|---|---|---|
| ʰ | h~ | ɦ |
| m | m | m |
| ɳ | N | ṇ |
| n | n | n |
| ŋ | q | ṅ |
| G~ | g̃ | |
| ʁ | G | ġ |
| ɠ | g? | g̀ |
| g | g | g |
| ʔ | ' | ' |
| ʕ | Q | q̇ |
| ɗ | d? | d̖ |
| dʼ | d! | d̗ |
| ǂ | d* | d̬ |
| d̚ | d. | ḏ |
| ɖ | D | ḍ |
| dˤ | dQ~ | d̰ |
| d | d | d |
| ɓ | b? | b̖ |
| bʼ | b! | b̗ |
| b | b | b |
| pʼ | p! | ṕ |
| ʘ | p* | p̂ |
| p̚ | p. | t̄ |
| p͈ | p@ | x̌ |
| p | p | p |
| ʈʼ | T! | ṭ̗ |
| ʈ | T | ṭ |
| tʼ | t! | t̗ |
| ǀ | t* | t̬ |
| tˤ | tQ~ | t̰ |
| t͈ | t@ | ṱ |
| t̚ | t. | ṯ |
| t | t | t |
| kʼ | k! | k̗ |
| k̚ | k. | ḵ |
| ǃ | k* | k̬ |
| qʼ | K! | ḳ̗ |
| q | K | ḳ |
| k | k | k |
| χʼ | H! | ḥ̗ |
| χ | H | ḥ |
| h! | ħ | |
| h | h | h |
| ʐ | J | ȷ̈ |
| ʒʼ | j! | j́ |
| ʒ | j | j |
| ɬʼ | S! | ṣ́ |
| sʼ | s! | ś |
| ɬ | S | ṣ |
| sˤ | sQ~ | s̰ |
| s͈ | s@ | s̭ |
| s | s | s |
| ɸ | F | f̣ |
| fʼ | f! | f̗ |
| f | f | f |
| ʋ | V | ṿ |
| v | v | v |
| zʼ | z! | ź |
| zˤ | zQ~ | z̰ |
| z | z | z |
| ɮʼ | Z! | ẓ́ |
| ɮ | Z | ẓ |
| ðˤ | CQ~ | c̣̃ |
| ð | C | c̣ |
| θˤ | cQ~ | c̃ |
| θ | c | c |
| ɭ | L | ḷ |
| ǁ | l* | l̬ |
| lˤ | lQ~ | l̰ |
| l | l | l |
| ɽ | R | ṛ |
| r | r | ṙ |
| ʃʼ | x! | x́ |
| ʂʼ | X! | x̣́ |
| ʂ | X | x̣ |
| ʃ͈ | x@ | x̭ |
| ʃ | x | x |
| ɰ | W | ẇ |
| wʼ | w! | ẃ |
| ʷ | w~ | ẉ |
| w | w | w |
| ʲ | y~ | ẏ |
| j | y | y |
These are all the combos for the letter a, same applies to all vowels.
| IPA | ascii | simplified |
|---|---|---|
| æ | A | ạ |
| œ | a$ | a̤ |
| ˈa | a^ | ȧ |
| aː | a_ | ā |
| aʼ | a! | a̱ |
| a̰ | a& | a̰ |
| a͈ | a@ | a̖ |
| a˦ | a+ | á |
| a˥ | a++ | a̋ |
| a˨ | a- | à |
| a˩ | a-- | ȁ |
| a˧˥ | a/ | ą́ |
| a˩˥ | a// | ą̋ |
| a˥˧ | a\ | ą̀ |
| a˥˩ | a\ | ą̏ |
| a˩˥˩ | a/\ | â |
| a˥˩˥ | a\/ | ǎ |
Note: Exact tone sequences can be represented with sequences like
a+a++a--, where one vowel is spread across multiple tones. But common
tones, across languages, can take advantage of the shortened
syntax/encoding.
npm install @termsurf/talkimport make from '@termsurf/talk'
make('aiyuQaK') // => 'aiyuq̇aḳ'@termsurf/text: Builds off Talk spec to convert different scripts/orthographies into TalkText.@termsurf/tone: Takes the Talk formatted ASCII text and renders it using ToneText.
Here we have included a system inspired by the Double Metaphone algorithm, which is an algorithm which creates a simplified pronunciation "hash" of some input text, usually English or other Indo-European languages.
Since Talk is itself a simplified ASCII pronunciation system for any of the world's languages (like X-SAMPA or IPA, but easier to write), it was straightforward to make a system where we progressively simplify the pronunciation from accurate to only simplified consonants and no vowels. There are 5 categories of things which get tinkered with when "refining" the pronunciation from its most accurate form, to the most basic form:
- vowel: none, one, basic, all. No vowels, the
avowel, the 5 basic vowelsi e a o u, or any possible vowel allowed by Talk. - consonant: all, simplified. All possible consonants allowed by Talk, or a simplified subset, where it basically merges bp, td, xj, fv, sz, and kg, and gets rid of any consonant variants like click consonants or stop/tense consonants (Korean).
- tone: yes, no. Whether or not we include tone markers (useful in Chinese).
- duration: yes, no. Whether or not we include duration markers (useful in Sanskrit).
- aspiration: yes, no. Whether or not we include aspiration markers (useful in Indian languages).
By combining all these characteristics, we end up with something like
this (for the word by~oph~am, which has palatalization, aspiration,
and a few vowels and non-simplified consonants):
import read from '@termsurf/talk/make/read'
const list = read('by~oph~am')
[
{
text: 'by~oph~am',
mass: 405,
load: {
consonant: 'all',
vowel: 'all',
tone: 'yes',
aspiration: 'yes',
duration: 'yes',
},
},
{
text: 'by~ph~m',
mass: 324,
load: {
consonant: 'all',
vowel: 'basic',
tone: 'yes',
aspiration: 'yes',
duration: 'yes',
},
},
{
text: 'by~opam',
mass: 270,
load: {
consonant: 'all',
vowel: 'all',
tone: 'yes',
aspiration: 'no',
duration: 'yes',
},
},
{
text: 'pyopham',
mass: 270,
load: {
consonant: 'simplified',
vowel: 'all',
tone: 'yes',
aspiration: 'yes',
duration: 'yes',
},
},
{
text: 'by~pm',
mass: 216,
load: {
consonant: 'all',
vowel: 'basic',
tone: 'yes',
aspiration: 'no',
duration: 'yes',
},
},
{
text: 'pyphm',
mass: 216,
load: {
consonant: 'simplified',
vowel: 'basic',
tone: 'yes',
aspiration: 'yes',
duration: 'yes',
},
},
]The mass is basically a "weight" for now, to say how many features it
included, i.e. how close to the actual pronunciation it was. The smaller
the mass, the less it is like the original pronunciation.
You then use the text as a key in a lookup table to find words
matching that refined text pronunciation. You likely will find the same
term in several spots, but you can just filter those at at query time.
That's about it! Now have to play with this in production to see how useful it is in practice for building pseudo-fuzzy dictionary search.
Using the library, you can also count the number of syllables in a word, and convert IPA text into ASCII Call Text.
import talk from '@termsurf/talk/make/talk'
import syllables from '@termsurf/talk/make/talk/syllables'
talk('kxɯʎʎikʰa̠da̠') // => 'kHOly~ly~ikh~a@da@'
syllables('kHOly~ly~ikh~a@da@') // => { size: 4 }import talkToIPA from '@termsurf/talk/make/talk/ipa'
import talkToXSampa from '@termsurf/talk/make/talk/xsampa'
import ipaToTalk from '@termsurf/talk/make/ipa/talk'You can also transform Talk into ToneText by writing it in ASCII, and running it through the tone text code, which is freely available and open source there.
import tone from '@termsurf/tone'
// make it for the font.
tone.make('a+a+si-kiri-imu-') // => 'a3a3si4kiri4imu4'MIT
This is being developed by the folks at TermSurf, a California-based project for helping humanity master information and computation. Find us on Twitter, LinkedIn, and Facebook. Check out our other GitHub projects as well!