A measured speaking language Φ
Tune is a spoken language where a word built out of other words can be read exactly one way. Every root is one syllable. Join two and the seam is never in doubt, so a listener cuts the word apart in the one place the writer put it together.
lif + vit = liflvit leaf life
tok + gan = toksgan tree garden
horn + lif = hornlif Takakia ceratophylla, the horn leaf moss
| sounds | 27: 22 consonants, 5 vowels, 3 diphthongs |
| roots | 4,096, every one a single syllable |
| legal forms the rules allow | 15,361 |
| forms that also stay apart in the ear | 7,625, so 3,529 spare |
| ordered pairs of roots | 16,777,216, of which 0 read two ways |
| seams that write a joining letter | 28.62%, the other 71.38% write nothing |
| mean length of a joined pair | 9.42 sounds |
Three things are being held at once, and they pull against each other.
- No compound is ambiguous. Not at two roots, and not at any depth. This is decided by a proof rather than by a spot check.
- Nothing sounds like anything else. No two roots of the same shape differ by a single soft sound, which is what throws away 7,736 of the legal forms.
- Words stay short. One syllable per root, and the shortest roots go to the concepts the most other concepts are built from.
Every figure on this page is measured, not estimated. The generated
quick guide is case/v24/readme.md, written by
case/v24/code/guide.ts from the one rule file,
case/v24/code/rule.ts.
| mark | sound | IPA | note |
|---|---|---|---|
m |
mark |
||
n |
note |
||
q |
sing |
ŋ | the -ng sound |
b |
band |
||
d |
deed |
||
g |
gift |
ɡ | |
p |
play |
||
t |
time |
||
k |
king |
||
h |
heal |
||
s |
soul |
||
z |
zone |
||
f |
fire |
||
v |
vibe |
||
x |
ship |
ʃ | the "sh" sound |
j |
beige |
ʒ | the "g" sound there |
c |
thor |
θ | voiceless "th" |
C |
this |
ð | voiced "th" |
y |
yard |
j | |
l |
love |
||
r |
rise |
with a spanish, arabic or indian accent | |
w |
wave |
Vowels are the spanish i e a o u. The three diphthongs are ai, au
and oi.
j is the one letter that moves twice. Tune's j is ʒ and Tune's
y is IPA's j, so a chain of replacements would send y to j and
that j on to ʒ. Every conversion walks the letters once instead.
w is a root letter again. It was held back through v17 to mark a
dropped sound at a seam. v24 marks that case with nothing, which returns
w to the alphabet along with the kw sw tw openings, and returns 17
words that could not be spelled without it: water, with, we,
what, why, walk, wait, wave.
Every root is one syllable. It opens on a consonant, holds one vowel or one diphthong, and closes on a consonant. Either end may carry a cluster.
CVC CVCC CVCCC CVVC CVVCC CVVCCC
CCVC CCVCC CCVCCC CCVVC CCVVCC CCVVCCC
CCCVC CCCVCC CCCVCCC CCCVVC CCCVVCC CCCVVCCC
A single consonant opens a root except q, and closes one except y,
w and h. The clusters are fixed lists.
starts 30
bl br cr dj dr fl fr gl
gr kl kr pl pr sk sl sm
sn sp st tr tx xl skl skr
spl spr str kw sw tw
ends 34
dj ft fk lc lf lk lp lt
mp nd nt qk rb rd rf rg
rk rp rt sk sp st tx zb
zd zg ndj ntx qkc gd kt rm
rn lv
no root opens wa
at most one of x and j
at most one of c and C
no il el ir er ul ur except -ul word finally
no same liquid across a vowel rar, lal refused
no medial h
nothing on the taboo list
no liquid either side of the vowel when BOTH ends are clusters
blark and blalk refused
The rules above say what is legal. This says what is worth using. mir
and nir are both legal and one of them has to go, or a listener cannot
tell them apart.
Two roots must be at least two steps apart, where one step is a near sound in one position. A single soft difference is the only thing forbidden, since a pair differing in two places is already clear.
at an onset s z f v x j c C
at a coda b d b p d t g k p t
f v f c v C c C l r
two nuclei never near, except ai oi, which close alike
Where a sound sits changes what it is near. A sibilant's place is
heard in the vowel that follows it, so s against x is one word twice
at the end of a root and two words at the front of one.
In a root carrying a cluster, four more pairs count as near at every consonant position.
b p d t g k m n
nain main no cluster both stand
brain prain cluster onset one of them goes
snain smain cluster onset one of them goes
faind vaind cluster coda one of them goes
CVC and CVVC are exempt from that tightening, which is what keeps
the short words. A cluster crowds the ear, and three sounds with nothing
else competing do not.
Roots run straight together, and 71.38% of seams write nothing at all. The rest write one letter, for one of two reasons: the two sounds smear, or the cut would be in doubt.
| example | written | the seam | share |
|---|---|---|---|
| rij + drom = rijdrom rhythm drum |
nothing | anything else | 71.38% |
| tok + gan = toksgan tree garden |
s |
two stops of one place, voiceless left | 2.98% |
| sid + tok = sidztok seed tree |
z |
two stops of one place, voiced left | above |
| mant + drom = mantsdrom mountain drum |
s or z |
the same, after a cluster coda | above |
| must + drom = mustldrom | a liquid | the same, after a coda opening on s |
above |
| lif + vit = liflvit leaf life |
a liquid | a fricative voicing pair | 1.90% |
| vit + tok = vitok life tree |
the sound once | the same sound doubled | 4.84% |
| red + drom = redldrom red drum |
a liquid, both roots whole | the same, cluster on the right | above |
| drom + man = dromzman drum mind |
z |
a doubled nasal | 0.27% |
| djul + lun = djulrlun jewel moon |
r after l, z after r |
a doubled liquid | 0.07% |
| ram + yam = ramlyam dark day |
a liquid | a root opening on y |
2.00% |
| mim + skliq = mimlskliq | a liquid | a three letter cluster at the seam | 15.91% |
| mim + pifk = mimlpifk | a liquid | the cut is in doubt | 0.63% |
| share | |
|---|---|
| for SOUND, the six cases | 12.07% |
| for the CUT, a three letter cluster | 15.91% |
| for the CUT, this pool would read two ways | 0.63% |
| anything at all | 28.62% |
| nothing written | 71.38% |
The two cut rows are not the same kind of fact. A three letter cluster takes a liquid because of what the two roots ARE, so it holds whatever words the language later gains. The last row is what THIS pool of 4,096 would otherwise spell two ways, so it moves when the words do.
Six of the rows above write "a liquid" rather than an l, because an
l stops being a joiner the moment it lands beside another one. ll is
one long l to a listener, and a joiner nobody can hear is not a
joiner. So the letter moves, and where both liquids are spoken for it
leaves the liquids entirely.
the coda is l and a consonant r balc + Cak = balcrCak
and the right root opens on r ri
l meeting l, l meeting r, r meeting l i
the left root closes on l r bal + yan = balryan
anything else l bat + yan = batlyan
The i breaks the il and ir rhyme ban on purpose. That rule
governs ROOTS, where a close vowel before a liquid is swallowed into it.
A seam is not inside a root, and a vowel is the one thing that cannot be
mistaken for cluster material, so it is the safest joiner the language
has and the least available inside a word.
Every joiner stands BETWEEN two roots, so both keep their spelling and
the joiner lifts back out. One row does not. A doubled sound is said
once and nothing marks it, so vit + tok is vitok. A VOWEL standing
where an opening consonant should be is what tells a reader a sound was
dropped, and which sound it was is fixed, since a twin is two of the
same consonant.
every ordered pair 16,777,216 read two ways 0
3 roots, 2,000 compounds read two ways 23
4 roots, 2,000 compounds read two ways 17
Pairs are exhaustive. The deeper runs are samples, because a disagreement can span three roots without appearing in any pair.
Two facts carry it. No root opens or closes on a vowel, so a vowel between two consonants is never root material. A joiner sits where the cluster lists allow no such letter, so it can be lifted back out.
The pair result is a proof and not a spot check. Brute force does not scale, since 4,096 roots is 68 billion triples, so the question is asked as whether the root set is a uniquely decodable code, which Sardinas and Patterson decide exactly and for every depth at once. It passes on the bare stream and on the stream where every root may also carry a joiner, which is more than the rule ever emits. The leftover fragments the search turns up all begin with a vowel or are single letters, so not one of them can be a root, and that argument does not depend on which 4,096 roots were chosen. The check calibrates itself against a code known to be ambiguous first, so a clean pass is not just a test that cannot fail.
v17 had a third fact and v24 spent it. No root held a w, so a w
was always a joiner and a dropped twin was marked. Writing the doubled
sound once and marking nothing buys w as a root letter and 17 pinned
words back, and costs the guarantee at depth three: zarg + gif and
zar + gif both spell zargif. Pairs still read one way. Deeper
compounds are the 23 above. Setting V24_TWIN_W=1 puts the w seam
back and takes w out of the alphabet again, for comparing the two.
| opening | legal | usable |
|---|---|---|
| one consonant | 6,458 | 3,281 |
| a two letter cluster | 7,445 | 3,526 |
| a three letter cluster | 1,458 | 818 |
| total | 15,361 | 7,625 |
Legal is what the sound rules allow. Usable is what survives the distance rule, and the gap is what that rule costs. The language spends 4,096 of the 7,625 and holds 3,529 spare. The per-shape table is in case/v24/readme.md, and the lists themselves are in case/v24/base/term/usable.
There are 4,096 base roots and everything else is a compound. The question is never how many concepts are needed. It is which 4,096, and what each one displaces.
A concept earns a root only by passing all five of these.
1 it is an IDEA, not a word form
2 it is not a proper name
3 it is not the source's own grammar
4 it cannot be said from bases already held
5 it is load bearing enough to be worth a root
One idea, not one word form. anger takes a root and angry does
not, create does and creation does not. Two roots on one point
spends two of 4,096 semantic coordinates on one meaning.
No proper names, ever. Not a person, not a place, not a people, not
a deity. A proper name is answered by DESCRIBING the thing. The test is
whether the meaning survives a speaker who has never heard of the
referent: narrow leaf passes anywhere, Smith's does not. The
taxonomy alone carries 65,596 eponyms and 3,700 toponyms, and every one
of them gets a description instead.
A closed set gets a head word and one trait each, never a root per
member. The 118 elements cost one root for atom plus concepts the
language already holds, so hydrogen is water atom and carbon is life
atom. The same shape covers planets, months, weekdays, colours, metals,
body parts and directions.
The fourth rule is why the set is grown rather than chosen. Whether
a meaning is sayable depends on what else is held, so adding fleece
turns goldenfleece from a base into a compound. The build runs to a
fixpoint for that reason, re-asking after every round what is still
unsayable.
A compound earns a stored form only when it is COMMON. One said twice is built fresh from its parts, like any language does.
Where the rules allow it, a root takes the sound of the English word for
its concept. A speaker who meets lif for leaf or drom for drum
already knows it, and that is the cheapest vocabulary a language will
ever get.
leaf lif drum drom mind man
dark dark light lait sound saund
English first, Sanskrit where English gives nothing. Not Chinese and not Arabic. Their sound systems are far enough from Tune's that the echo does not survive the transcription, and a borrowed form that no longer sounds like its source has bought nothing.
Chinese is the model for compounding, not for sound. It builds a technical vocabulary out of common concrete morphemes and almost never borrows, which is the constraint Tune is under.
电脑 electric + brain computer
火山 fire + mountain volcano
长颈鹿 long + neck + deer giraffe
Three traits decide a term, ranked because they conflict: minimalism, then accuracy, then obviousness. A guessable name that is subtly wrong poisons every compound built on it, so accuracy outranks obviousness. A compound needing four parts is usually a sign the base set is missing something.
person heads 358 other concepts, place 257, part 255. A word a
hundred other words lean on is said inside every one of those
definitions, so its cost is multiplied rather than counted. Three
sounds are the scarcest thing the language has, and they belong to the
core concepts, the pins and the relations, not to whatever a domain
corpus happens to repeat.
Ranked above frequency, which is the weakest signal and the loudest:
PRODUCTIVITY how many other concepts are built out of it
USEFULNESS what becomes sayable the moment it is held
FREQUENCY how often anybody happens to say it
tooth is said seven times in the English list and blocks 543 species.
stone is said thirty six times and blocks 647. Neither survives a
ranking by frequency and neither can be lived without.
There is one joining mechanism and derivation uses it. An affix is a root, joined at a seam by the same rule as any other root.
yod past reC ongoing kif future
zek one who does kan capable of lef less of
ved study of tul it is a tool huf opposite
58 of them are written so far, in case/v24/base/term/affix.csv. The commonest grammar must be the cheapest to say, because it lands on every other word.
Said aloud, a noun takes the ending -a, and the stress falls on
the second to last vowel. rij + drom is spoken rijdroma,
/ɾiʒdɾˈoma/.
A liquid joiner takes a helper vowel in speech and never in writing,
or a voice runs the three consonants together and the joiner is not
heard. djulrluna is said djulırluna, /dʒulɯɾlˈuna/, where ı is ɯ.
| path | what |
|---|---|
| case/v24/readme.md | the generated quick guide, every figure measured at generation time |
| case/v24/design.md | the rules the vocabulary work obeys, the meaning side |
| case/v24/process.md | how the base set is built, step by step |
| case/v24/code/rule.ts | the sounds, the shapes and the seam, held ONCE, read by both the writer and the reader |
case/v24/base/term/usable/ |
every usable root, one file per shape, plus all.txt and all-ipa.txt |
case/v24/base/term/legal/ |
every root the sound rules allow, before the distance rule |
case/v24/base/term/form.csv |
every seated concept, with its root, its length, and why it holds that one |
case/v24/base/term/pinned.csv |
478 concepts whose form never moves |
case/v24/base/term/affix.csv |
the affix roots |
case/v24/base/term/species.csv |
39,640 species, and the Tune name for each one that has one |
case/v17/base/voice/ |
the recordings, and the video in both shapes |
| case/readme.md | every earlier version, and what each was for |
pnpm --dir deck/tune v24:every the root lists
pnpm --dir deck/tune v24:pinned the pins
pnpm --dir deck/tune v24:words the words file
pnpm --dir deck/tune v24:guide the quick guide
pnpm --dir deck/tune v24:test the suite
The sound system is settled and measured. The vocabulary is not finished, and the numbers that say so are kept beside the ones that look good. Full list in case/v24/todo.md.
39,640 accepted species
26,223 with any Tune name at all
16,700 of those sharing a name with another species
3,300 with no literal meaning known for either word
The largest cause is one broken link in the loop: the count of species a missing concept blocks is written out and nothing reads it back, so a concept blocking 741 species competes on its gloss frequency instead. Every figure above is plants and animals. Genes, minerals, medicines, tools and anatomy are untouched, so none of it should be quoted as covering a terminology in general.
These are the parts your mouth and tongue shape.
Sharing yourself in new ways to the lake.
These are the symbols representing sound bits.
Combining into a rhythm like the drum hits.
Soon you'll notice that each twist and turn fits.
These are the wholes that link the speech tones.
Morphing our intelligence as we play with each stone.
These are the words we separate with spaces.
Defining them carefully so we get down to the basics.
You'll soon learn how to use them in statements.
These are the threads that frame what you speak.
Giving others a peek at your inner geek.
These are the sentences your voice brings in the night.
Manifesting the invisible in the mind's eye.
Now you're a chat master, next is an invite.
These are the structures that make the light whiter.
Making it possible to communicate about the higher.
These are the trees in which we mold information.
Seeing ourselves and the universe as one big computation.
The key thing to remember is patterns and patience.
These are the networks merging everything into one.
Connecting knowledge together like the moon and the sun.
These are the primary thing the earth yields.
Filling the memory with energy from brook and field.
If you've made it this far it's the stone that you wield.
Note: Tune is in the prototype phases right now. Check out the website for the latest grammar, lexicon, and other things. And a recent spreadsheet too.
Copyright 2021-2025 ClueSurf
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
Made by ClueSurf, meditating on the universe ¤. Follow the work on YouTube, X, Instagram, Substack, Facebook, and LinkedIn, and browse more of our open-source work here on GitHub.