Faleni Language Specification (v0.4)

This document fully defines Faleni. If something isn't here, it follows from the rules here; there are no exceptions hiding elsewhere.

1. Design goals (and the trade‑offs we chose)

The brief: the most optimal and efficient language, easy to learn, easy to apply, with no rules that don't make sense, a system so there's a word for everything, and completely original.

"Optimal", "easy", and "original" pull against each other, so we made the choices explicit:

The non‑negotiables: no irregularities, no silent rules, one sound per letter, every word invariant in shape, an open‑ended but collision‑free way to coin words, and nothing borrowed from an existing language.

2. Sounds and the alphabet

Faleni uses 16 Latin letters, each with exactly one sound. No digraphs, no silent letters, no capitals required.

Vowels (5)

a e i o u: as in father, bet, machine, more, flute. Length and pitch never change meaning.

Consonants (11)

p t k f s h m n l w j

• j = English y (as in yes). ju is pronounced "you".
• Accent tolerance: voicing is not meaningful, so p≈b, t≈d, k≈g, f≈v, s≈z are all accepted pronunciations of the same letters. A speaker who voices a stop is still understood. This lets speakers of many native languages be understood without retraining their mouths.

Left out on purpose

No r (varies wildly; clashes with l); no th, no tones, no vowel‑length or stress contrasts; no voiced/voiceless pairs competing for meaning. That gives 11 × 5 = 55 simple, well‑spaced syllables.

3. The onset‑family architecture (Faleni's signature feature)

Because every root is invented, we gave the vocabulary a backbone: the first sound of a content word marks its meaning‑domain.

Closed sets sit outside this scheme (they're learned once as a group):

• h and j → grammar particles and question words (ha is, he not, jo plural, jeta what…)
• w → pronouns and deictics (wa I, we you, wi they, wo this)
• numerals → so ta ne ki fo pe lu si mo ku (0–9) and na to fu (10/100/1000)

What the scheme costs. Fixing the onset buys navigability and pays for it in capacity. Each family has 5 x 11 x 5 = 275 two-syllable shapes, and the rule "no two words in a family one sound apart" burns roughly 18 shapes per root, so a family saturates at somewhere under a hundred roots. k-, l- and n- are already there (python3 tools/faleni.py capacity prints the current state). That is a real limit, and it is the limit the compounding engine (§8) exists to absorb: when a family is full, you build the word instead of coining it, which is what the design says to do first anyway. Three-syllable roots remain available for anything genuinely atomic.

Why only the first consonant? The 1600s "philosophical languages" (Wilkins) encoded a word's full taxonomy into its letters, so salmon and trout differed by one sound and were impossible to tell apart. We took the lesson: fix only the broad domain (one consonant), and let the rest of the word vary freely for distinctness. The onset is a memory hook and a coinage recipe, never a rule you apply while speaking.

4. Word shape, stress, and self‑segregation

• Syllable = (C)V. Optional consonant + vowel. The only syllable type. No clusters, no final consonants.
• A word is one or more syllables: wa, tuna, jeta, faleni.
• Stress is always on the first syllable. Never written, never contrastive; it just marks where each word begins, so speech is self‑segregating.

Word length tracks frequency (Zipf's law of abbreviation):

5. The whole grammar, in 10 rules

1. Read it as written. One letter, one sound. j = "y". Stress the first syllable.
2. Sentence = Subject ha Predicate. The word ha ("is/does") marks where the subject ends and the predicate begins. wa ha fa = "I speak."
3. Head first, then its stuff. Verb before object: wa ha mu we = "I love you." This one ordering rule is used everywhere (rule 4).
4. Modifiers follow what they modify. Adjective after noun, adverb after verb, number after noun, owner after the owned. tu le = "big house"; kali liwa = "walk fast"; tu wa = "my house".
5. Words never change shape. No endings, ever. kema is eat / ate / will eat / eating.
6. Roots have no fixed part of speech. fa = speech (noun), to speak (verb), verbal (modifier) depending on slot.
7. Negate with he, right before what you deny. wa ha he fa = "I don't speak."
8. Ask yes/no with ho at the end. Ask everything else with a je‑ word (jeta what, jepa who, jewa where, jeni when, jesa why, jefe how, jenu how‑many), placed where the answer would go. we ha ka jewa? = "Where are you going?"
9. Join with hi (and) / hu (or). wa hi we = "you and I".
10. Make new words by describing them, head first (§7). That's the whole vocabulary system.

There is no rule 11.

All ten, plus the sounds, the families and the core words, fit on one printable page: the cheat sheet.

6. Pronouns, plural, time

Pronouns (one set, no gendered or object forms, all start w):

Plural is the word jo ("many") after the noun: tu jo = houses, wa jo = we, wi jo = they. Possession needs no extra word, the owner follows: tu wa = "my house" (optional ja for clarity: tu ja wa).

Tense is optional. Faleni is tenseless by default; context carries time. When needed, add a time word, napo (past), nima (now), nase (future):

• wa ha kema napo = "I ate" (lit. "I eat past").
• wa ha ka nase = "I will go."

7. Numbers

Base‑10, perfectly regular. Digits are one syllable; powers of ten are their own short words.

so0 ta1 ne2 ki3 fo4 pe5 lu6 si7 mo8 ku9 na10 to100 fu1000

Build a number big‑part first, summing the pieces:

• ne na ki = 2·10 + 3 = 23
• ki to = 3·100 = 300
• ne fu · ne na pe = 2000 + 25 = 2025

Numbers are modifiers, so they follow the noun: tu ki = "three houses".

Decimals, fractions, ordinals, negatives, dates, and clock time are all specified in Numbers & extras.

8. Building words, so there is a word for everything

You almost never memorize a new word; you build it. Four layers, in order of preference (full protocol in Coining words):

Layer 1: Core roots

The lexicon (the dictionary): ~500 concepts, each in its onset‑family. These are the atoms.

Layer 2: Compounds (the main engine)

Combine roots head first: the first root is what it is, the rest narrow it:

Layer 3: Semantic heads (regular derivation)

A few roots are conventionally used as "heads", no special affixes to learn:

Layer 4: Borrowing (proper nouns & untranslatables)

Transliterate into Faleni sounds and mark with ji (the name marker), so the word is never mistaken for a compound:

• Tokyo → ji tokijo · Claude → ji kolote · America → ji amelika

Transliteration: map each foreign sound to the nearest Faleni letter (r→l, v→f/w, g→k, b→p, d→t, th→t/s); break clusters and final consonants with a vowel (default o/i); drop sounds Faleni lacks.

Preference order: compound > derivation > borrowing. Borrow only for proper names or genuinely arbitrary terms.

9. Why this is "efficient" in the information‑theory sense

• Channel‑robust: wide phonetic spacing + self‑segregating (C)V + first‑ syllable stress = high resistance to mishearing. We did not chase maximal compression, Shannon tells us an over‑compressed code has no error margin.
• Low learning entropy: few exception‑free rules ⇒ the whole grammar has a tiny description length. That is "easy to learn".
• Productive, not memorized: meaning composes, so the effective vocabulary is far larger than the root count. ~500 roots already express thousands of ideas.
• Effort tracks frequency: common things are short, rare things are long, so average message length is near‑minimal for real usage.

10. Status & roadmap

v0.4 (now): sounds, the onset‑family architecture, grammar, numbers, derivation, borrowing, a ~500‑root / 554‑entry a priori lexicon, a validator, a two‑way translator, ten lessons and two reference texts. Name verified not to be an existing language.

Natural next steps:

1. Grow the vocabulary, but mostly as compounds. The original target of ~600 roots no longer holds: k-, l- and n- have no clean two-syllable form left, and s- and m- are nearly there (python3 tools/faleni.py capacity). A family holds 275 two-syllable shapes and each root rules out about 18 of them, so saturation arrives under a hundred roots. That is the compounding engine's purpose, not a problem: build the word, and keep the compounds page growing instead.
2. ~~Fix the number tail: decimals, fractions, ordinals, dates, clock time.~~ Done in Numbers & extras, taught in Lesson 10, and handled by the translator.
3. ~~Optional aspect markers (started / ongoing / finished) as opt‑in particles, kept out of the core on purpose.~~ Done: ju marks an action in progress, and kipa / kine carry start and finish as ordinary verbs.
4. A Faleni ⇄ English translator / generator built on lexicon.csv.
5. A reference text (a translated short story) to stress‑test the design.

Everything above is additive; none of it changes the 10 core rules.