Grammar recovered from the Marain community lessons

This document prepares the conlang for further expansion and describes the deliberately limited phrase assistant on /marain/. It compares the 15 supplied lesson images with the 430-entry dictionary in KO0I/KO0I.github.io, commit bf89b578997c2466dc54b3dcca723935f2a6bccc.

The grammar is community material, not a claim that Banks published a complete Marain grammar. Evidence is labelled explicit (a stated rule), attested (an example), inferred (a cautious generalisation), or unresolved. New output assembled from rules is labelled a draft even when every word is known.

1. Representation and spelling

Explicit: Use the revised romanisation in IMG_2187: ah → a, oh → o, ee → i, eh → e, je → ye, oo → u. Preserve uh, ih, and ay. Native glyph keys remain the case-sensitive internal alphabet already used by dict.js and alpha.js; they are not ordinary English letters.

Use apostrophes to disambiguate letter boundaries when needed: ra'yuh, ra'ye, sein'gaf, s'ha, and g'nayt. These separators are not separate sounds. Normalise via glyph keys when comparing dictionary words, not by indiscriminately deleting letters or lowercasing the keys.

Correction to the initial OCR transcript: Lesson 16 clearly prints hanggra, not hangra, when enlarged. Its glyphs are haGgra (hang + gra). The dictionary addition and assistant preserve that spelling.

Unresolved: The first lesson describes its writing as right-to-left, but the subsequent word tables align with the site’s existing left-to-right glyph sequences. Do not reverse strings globally or reinterpret the font encoding from that one sentence.

2. Cases and clause roles

Explicit, Lesson 1 (IMG_2186):

Role	Roman suffix	Internal suffix	Example
Nominative: subject/actor	-uh; -yuh after a vowel	u; yu	pren → prenuh; ra → ra’yuh
Accusative: object	-va	va	maraynva
Dative: recipient	-vihl	vil	gevihl
Originative: source/origin	-sa	sa	raybihnsa
Locative: location	-li	lE	prenli
Comitative: companion/association	-ye	j	ra’ye
Vocative/absolutive	none	none	a bare name or citation form

The source calls the dative a “direct object,” but its example is “to them.” The assistant follows the example and reserves accusative for the direct object. This discrepancy is documented rather than converted into two competing object rules.

Explicit: Case supports flexible word order. Inferred implementation convention: use subject–verb–object when generating an ordinary draft. Source examples also put verbs or objects first; the convention is not a new grammatical requirement.

Attested: ra'yuh gore ivemiva dam ge — I meet your friend. ra'yuh kabo maraynva — I speak Marain. Bare verbs occur with English present and past translations. Do not invent English-like person endings, past suffixes, future auxiliaries, or obligatory plural agreement.

Unresolved: Coordinated noun phrases, subordinate clauses, voice, predicate complements, and case scope across complex phrases. The first assistant handles a restricted clause, one direct object, and at most one explicit recipient/location/origin/companion phrase. Unsupported structures fall back to lexical matches.

3. Pronouns

Person	Singular	Plural	Evidence
First	ra	ora	Lesson 1
Second	ge	llay	Lesson 1
Third	to	wuh	Lesson 1
Fourth/impersonal	ong	onge (Lesson 1); ongi (dictionary)	Variant conflict

English he/she/it map to to; the evidence does not require gendered forms. The UI makes its defaults explicit: “you” is singular and “they” plural. “You all” selects llay. Singular “they” needs future disambiguation rather than guessing from context.

onge and ongi encode different final vowels. Record the older form as a source variant; do not count it as an unrelated new meaning or silently replace the current dictionary’s ongi.

4. Possession and noun association

Explicit, Lesson 21 (IMG_2198): English “I have a flower” is yesayn lomra'yuh ra'ye: literally, a flower exists with me.

Generalisation: yesayn + thing.NOM + holder.COM.

• yaf means physically hold, not abstract possession.
• bahllavecht means hoard, own more than one can use; it is not a neutral translation of “have.”
• Keep these meanings distinct. Do not apply the possession template to English auxiliaries such as “I have eaten.”

Attested, Lesson 20: ivemiva dam ge — your friend. Noun association uses noun + dam + possessor.

Explicit, Lesson 11: da replaces dam before a word starting with the Marain letter named ma (the m glyph). The initial transcription’s wording could be read as the syllable ma; the image identifies the letter, and the current dictionary specifies initial m. The implementation applies this to a following initial m glyph.

5. Questions, negation, and modification

Explicit/attested, Lesson 16: hanggra + clause asks whether the clause is true. Example: hanggra kabo geyuh maraynva — Do you speak Marain?

Attested, Lesson 7: yokayshay sayno geyuh — How are you doing? shay sein'gafva datsha gevihl — What is your name?

Explicit, Lesson 1: shay = who/what/which; yokay makes an adjective adverbial. Attested, Lesson 12: yokay ivemi conveys “like a friend,” so the example permits a nominal base too. Do not declare that every English -ly adverb has a mechanically corresponding Marain form.

Attested, Lesson 15: x lig y = more y than x. This supports comparative constructions but not an unrestricted superlative or degree system.

Explicit/attested: hech covers no/not/without. ra'yuh nadek hech ubiva is translated “I don’t do drugs.” Unresolved: a general negation position and scope rule. The assistant knows this attested negative sentence but falls back for new negative clauses instead of moving hech around arbitrarily.

Unresolved: obligatory adjective position and agreement. The optional adjective + head order in generated simple noun phrases is provisional and visibly labelled in the interpretation notes. English articles are omitted in covered noun phrases, following the examples; this does not assert that definiteness is impossible in the language.

6. Derivation and semantic construction

Explicit, Lesson 7: verb → noun with -i after a consonant and -wi after a vowel. Examples: nadek → nadeki, kabo → kabowi.

Attested: Existing dictionary nouns can denote an action, product, or agent (gafmarawi, insleiltsimarawi, etc.). The productive suffix alone does not settle which sense a new coinage has. A coined derivation needs a gloss and provenance.

Explicit, position sheet IMG_2188: o- turns the listed spatial terms into verbs: g'nayt → og'nayt, “around” → “to orbit.” This evidence licenses that family, not arbitrary prefixing to all nouns.

Explicit, Lesson 1: ray- indicates uniqueness. Attested: bihn → raybihn (culture → the Culture), muhrer → raymuhrer (mind → Mind). Do not translate every English “the” into ray-.

Attested: Compounds include guhchetpren (land vehicle), choldicheinda (animal matter/meat), and sein'gaf (identity-word/name). Inferred: compounding is available for expansion. Unresolved: universal head direction, linking vowels, and semantic transparency. Record whole compounds rather than splitting arbitrary words and claiming their meanings are compositional.

Explicit, Lesson 1: pe-X-t means “seem to X”; perat is pe + ra + t, not the bare verb to be. This prevents subjective impressions from becoming false identity statements.

7. Time and number

Attested, Lesson 12: bechumli, ikili, beyanli mean in the past, present, future. These are locative forms of time nouns, not demonstrated tense suffixes. The assistant does not fabricate aspect or perfect-tense rules.

Explicit, IMG_2189 and Lesson 18: numbers are base eight. Digit words are nhech sto hre dosa llamih ko gol lyeway. Decimal 8 → stonhech; 9 → stosto; 10 → stohre. Convert the value before writing Marain numeral sequences; don’t merely replace decimal digit names.

Attested: dosa dam uprayheva — three humans (object). Numeral + dam precedes the case-marked noun.

Explicit: amnwe = eight to the power of; mathematical tswa = times; uh = plus/and. Lesson 18’s examples include amnweko (8⁵), and golstoko dam tswa amnwehre uh dosa (615₈ × 8² + 3). Arbitrary expression parsing is outside the first assistant.

Unresolved: the Lesson 18 English heading says “4 or more digits,” while the corresponding glyph line appears to express “more than 3.” Treat this as a likely equivalent threshold formulation, not a new numeral value.

8. Vocabulary reconciliation

The comparison is based on the glyph alphabet, not raw romanisation string equality. This avoids counting g'nayt/gnayt, s'ha/sha, and ordinary case forms as new headwords.

Eleven added entries (430 → 441)

Romanisation	Internal key	Meaning	Source
nayla	nAla	away; opposite direction	IMG_2188 position words
sein	seEn	identity	IMG_2190, Lesson 7
sein’gaf	seEngaf	name	IMG_2190, Lesson 7
datsha	datSa	to call; name	IMG_2190, Lesson 7
metwuhshay	metwuSA	how many; how much	IMG_2194, Lesson 15
hang	haG	false; incorrect	IMG_2195, Lesson 16
gra	gra	true; correct	IMG_2195, Lesson 16
hanggra	haGgra	is it true that	IMG_2195, Lesson 16
hechkwemrawi	heHkwemrawE	meatfucker; taboo for a ship reading minds without permission	IMG_2195, Lesson 16
choldicheinda	HoldEHeEnda	meat; animal matter	IMG_2195, Lesson 16
kra	kra	all	IMG_2197, Lesson 20 gloss

Each added entry records its source and community-lesson status in dict.js. No unattested vocabulary is coined in this change. ray-, case suffixes, and productive number sequences are documented rules, not duplicate lexical entries. Proper names in examples are not promoted to generic vocabulary.

Existing definitions reconciled

Entry	Change	Evidence
yaf	Remove neutral “to have”; retain physical holding	Lesson 21 explicitly distinguishes them
insleiltsimarawi	Include work as well as worker	Lesson 20 table and example
dafihnye	Include new as well as young	Lesson 11
heldyet	Include organisation as well as structure/building	Lesson 12
tswa	Record mathematical times as a contextual sense	Lesson 18
muhrer	Include consciousness and software	Lesson 12
muhksih	Preserve dictionary death and explicitly record lesson to die as an unresolved noun/verb difference	IMG_2193 handwriting; glyphs support muhksih

The handwriting spellings chakllich, chaytscha, and spetshihch already match dictionary keys. They do not need duplicate entries. The original handwritten romanisation of “to die” remains uncertain; it must not establish a competing muksih headword.

How to extend this work

1. Add an attested lemma with its case-sensitive glyph key, part of speech, exact English aliases, source image/lesson, and status. Prefer clear single senses over substring matches from prose definitions.
2. Distinguish a genuinely absent lemma, a missing sense, a spelling variant, a regular inflection, and a conjectural derivation.
3. Add a sentence rule only with a supported pattern and examples. Test unknown subjects, verbs, objects, number, ambiguity, and scope; do not erase unmatched words to make output look complete.
4. Label new coinages and inferred grammar as proposals. Keep them out of automatic authoritative output until accepted.
5. Future work needs explicit decisions on singular they, tense/aspect, negation scope, adjective agreement, complex clauses, names, and when a noun-derived action/agent sense is intended.

9. Implementation limits and verification

The phrase assistant runs entirely in the browser. It uses longest exact English phrase matching, small explicit inflection lists, case morphology, and a small clause recogniser. It performs no network inference and uses no external model. Unknown vocabulary and genuinely ambiguous lexical matches appear literally as <??>, with the English span listed underneath.

For unsupported grammar it returns lexical fragments, labelled Partial match, instead of asserting that those fragments form a correct sentence. Known long phrases win over their components. The glyph view renders placeholders in a normal text font rather than trying to translate the placeholder into Marain glyphs. User text is inserted with textContent, not executable HTML.

Validation command: node --test tests/marain-phrases.test.cjs. Static integration checks validate unique IDs, labels, script order, and escaped placeholders. A full Jekyll build and visual browser checks were not available in the execution environment (Ruby/Jekyll and a Chromium binary are absent). These remain review checks before merging.

9. Community grammar extension v0.1 — adopted as proposed vocabulary

New work, not recovered canon (21 October 2026). This section is separate from the lesson-derived rules above. The existing dictionary already contains the attested Lesson 16 words hang, gra, and hanggra; no duplicates are added.

Seven proposed dictionary entries: ke (introduce a verbal complement); funuft (able to); gihwinuft (permitted to); shuhztnuft (should); kwemraku'ha (obliged to); wolugnuft (intend to); hangtcha (counterfactual if). The existing anong receives a proposed sentence-level maybe/might use, and hech receives an explicitly proposed rule for predicate-local negation.

The proposed linker ke introduces a verb complement. Its unexpressed actor inherits the matrix subject; an explicit nominative actor overrides that. A modal negated by hech means something different from its complement negated by hech. The source lessons do not document this full generalisation.

Existing temporal locatives express past/present/future without imposing new tense conjugations. hanggra questions remain recovered grammar. The phrase translator now supports a bounded subset of the proposed modal clauses; unsupported conditionals still receive partial results.

See the full extension for provenance, example sentences, and remaining limitations.

10. Community grammar extension v0.2 — hypothetical predicates and clause chaining

New proposal (2026-10-21), superseding the v0.1 requirement for ke. The earlier section documents the first extension; it is not proof that ke appeared in the lessons. Predicates now chain directly; ke becomes an optional boundary marker for ambiguous complements. An omitted subordinate subject inherits the controlling subject, and an explicit nominative can introduce a new subordinate actor.

New irrealis marker: 'tcha, appended to a predicate after dropping one final romanised ch, c, p, d, t, or y (check ch first; no repeated deletion). Apostrophe is a written suffix boundary, not a glyph. Examples: uprahay → upraha'tcha, funuft → funuf'tcha, okehi → okehi'tcha. The independent tcha still means if. The glyph-key spelling for upraha'tcha is OprahaCa, distinct from the original OprahA.

Belief predicate: heth is existing believe (not tsha, think about). Proposed nested belief construction: ra'yuh heth geyuh shuhztnuft uprahay nayla (“I think you should leave”). ra'yuh heth'tcha ... conveys hypothetical belief. ra'yuh okehi'tcha uprahay nayla conveys “I would like to leave.”

This remains a community extension, not Banks canon. The full v0.2 specification documents composition, examples, negation scope, and unresolved cases.

11. Community grammar extension v0.3 — patientive, resultative, prospective

New, proposed (2026-10-21): This section is not additional evidence from the recovered lessons. The earlier seven nominal cases remain attested. The extension adds an eighth nominal patientive case -vuh for an affected grammatical subject with unspecified agent, plus two predicate-level suffixes:

• 'inayng marks a consequent predicate, relating it to a preceding proposition as the speaker’s asserted consequence. It does not encode past tense or completion.
• 'wolug marks an intended, scheduled, or planned event not yet realized relative to its reference time. It does not imply requirement or guarantee completion. Existing kwemraku'ha independently marks obligation, and beyanli is optional future time.
• The new suffixes retain complete predicate stems; only hypothetical 'tcha uses v0.2’s ending-deletion rule. Apostrophes are orthographic boundaries rather than glyphs. Suffix order and stacking on one predicate remain unspecified.

Examples: tovuh tshainsle bechumli = “it was imagined”; ra'yuh uprahay'wolug nayla = “I plan to leave”; ra'yuh kwemraku'ha uprahay'wolug nayla = “I am scheduled and obliged to leave.” Three proposed lexical compounds are gralemih (know), tshainsle (imagine), and independent inayng (therefore), all tagged as proposed-extension in the current dictionary.

The current full specification maintains a strict separation between recovered forms and invented vocabulary and grammar.

12. Proposed poetic register v0.4

New, not recovered (2026-10-21). The earlier grammar is retained. Prospective and resultative predicates may optionally use poetic endings -a or -i without changing meaning: uprahay'woluga / uprahay'wolugi and kabo'inaynga / kabo'inayngi. The hypothetical 'tcha rule retains its original deletion behavior. Suffix stacking on a single predicate is still unresolved.

A deliberately limited Gaelic-inspired initial mutation follows po (“small”) only in the opt-in poetic register, with mappings p→f, b→v, t→th, s→sh, k→ch. It is reversible only with the contextual trigger and only when the mutated word does not collide with an existing headword. Prose is not automatically mutated.

Five clearly labeled proposals extend the vocabulary: lemihgra (test/verify), gaflemih (read), saynowetsiki (still), hechrayspatsmaychrayti (non-war/non-military, an approximation), and burach (mess/shambles, adapted from Scottish Gaelic bùrach). Full ship names and surface mutations are not dictionary headwords.

The poetic register spec distinguish recovered language from authored extensions and partial translations.