AMR AI Architecture — Letter / Root / Tasrif

AA — Allah's Arabic (ORIG1)

The divine language of the Qur'an, taught to Adam (AS)

28 letters | 1,710 roots | 77,877 Quranic tokens

Source: Qur'an (absolute) | amr_alphabet.py | amr_tasrif.py

BI — Bitig (ORIG2)

From Yafith (AS). Kashgari's Diwan (1072 CE) = primary attestation

26 phonemes | 1,900 roots (bitig) | 2,083 diwan entries

Source: Arabic MS (absolute) | amr_bitig_tasrif.py | amr_bitig_alphabet.py

▼

Layer 0 — Letters (the atoms)

AA LETTERS — amr_alphabet.py (29 entries)

Each letter has 11 fields:

Field	What it is	Example (د)
name	Arabic name	دَال
abjad	Numerical value	4
core_concept	Semantic atom	KNOCK
phonetic.makhraj	Articulation point	Tongue tip → upper teeth
behaviour.as_R1	When opens root	CALLING/POSITIONING
behaviour.as_R2	When core of root	WEIGHT/DEFINITE IMPACT
behaviour.as_R3	When closes root	ANCHORING
shifts	Downstream phonetic shifts	د→d, د→t (devoicing)
quranic_stats	Token/root counts per position	5,204 tokens, 223 roots
semantic_tendency	Full semantic description	KNOCK. POSITION. ACCOUNTABILITY.
paired_letters	Common letter combinations	د+ع = calling from depth

BI PHONEMES — amr_bitig_alphabet.py (26 entries)

Vowel harmony system:

Category	Phonemes	Function
BACK vowels	a, ı, o, u	Govern suffix selection (back set)
FRONT vowels	e, i, ö, ü	Govern suffix selection (front set)
Consonants	18 consonants	Root structure carriers
Harmony pairs	a↔e, ı↔i, o↔ö, u↔ü	Suffix variant selection

Violation of vowel harmony = contamination signal
BG10 (Vowel Harmony) is the structural law of BI

▼

Layer 1 — Roots (letters combine into meaning units)

AA ROOTS — compute_root_meaning()

Root = 2-4 letters. Each letter's core_concept combines:

Root	Letters	Semantic Core	Abjad	Tokens
د-ب-ب	KNOCK+CLOSURE+CLOSURE	Heavy crawling thing	8	14
م-ل-ك	ENCLOSURE+CONNECTION+TAP	Sovereign containment	90	185
ح-ك-م	CONSTRICTION+TAP+ENCLOSURE	Judging/binding decree	68	195
أ-ج-ر	STOP+PRESS+MOVEMENT	Compensation/wage	204	98
ج-د-ر	PRESS+KNOCK+MOVEMENT	Wall/enclosure	207	4

1,710 distinct roots | every root traces to 28 letters

BI ROOTS — diwan_roots + bitig_a1_entries

BI roots = typically 1-3 consonants (agglutinating, not inflecting):

BI Root	Headword	Kashgari AA Gloss	AA Root Traced
b-k	بكو (buko)	الملك والحاكم والعاقل	م-ل-ك / ح-ك-م / ع-ق-ل
b-z	باز (baz)	أُجْرَة الأَجِير	أ-ج-ر
q-n	قان (qan)	الدَّم	د-م-ي
t-m	تام (tam)	الجِدار	ج-د-ر
s-n	سان (san)	العَدَد	ع-د-د

1,900 bitig roots | 2,083 diwan entries | 1,242 distinct diwan roots

▼

Layer 2 — Tasrif: Consonant Structure (what letters are added to the root)

AA VERB STRUCTURE — 8 codes (5,862 tokens)

Code	Structure	Function	Tokens
HAMZA_PREFIX	أ prefixed	CAUSE action outward	2,097
DD_MIDDLE	R2 doubled	INTENSIFY from within	1,437
T_INFIX	ت after R1	Action turned INWARD	775
ALIF_INFIX	ا between R1R2	MUTUAL engagement	451
N_PREFIX	ن prefixed	PASSIVE reception	402
T_PREFIX	ت prefixed	Self-REFLEXIVE	370
ALIF_PREFIX_K	اسـت prefixed	SEEK root action	268
T_DD_COMBO	ت + R2 doubled	Self-intensification	62

BI VERB STRUCTURE — 8 codes (BV01-BV08)

Code	Suffix	Function	Example
BV01	-gan/-kan	PAST PARTICIPLE → noun	korgan (seen/seer)
BV02	-ur/-ar	AORIST/present	korur (he sees)
BV03	-di/-ti	DEFINITE past	kordi (he saw)
BV04	-maq/-mak	INFINITIVE	kormaq (to see)
BV05	-mish/-mush	EVIDENTIAL past	kormish (apparently saw)
BV06	-sa/-se	CONDITIONAL	korsa (if he sees)
BV07	-gali/-kali	PURPOSE	korgali (in order to see)
BV08	-a/-e	OPTATIVE	kora (may he see)

AA NOUN STRUCTURE — 12 codes (39,593 tokens)

Code	Structure	Function	Tokens
BASE	Pure root	Direct naming	32,006
Y_BETWEEN_R2R3	ي between R2R3	PERSISTENT doer	2,154
A_BETWEEN_R2R3	ا between R2R3	Extended STATE	1,343
A_BETWEEN_R1R2	ا between R1R2	Agent/actor	966
M_PREFIX	م prefixed	PLACE/TOOL of action	1,413
M_PREFIX_W	م prefix + و	PASSIVE participle	664
+ 6 more	T_PREFIX, HAMZA_PREFIX, DD_R2, etc.

BI NOUN STRUCTURE — 6 codes (BN01-BN06)

Code	Suffix	Function	Count
BN01	-liq/-lik	ABSTRACT noun / quality	88 hw
BN02	-ci/-chi	AGENT (doer)	-
BN03	-man/-men	AUGMENTATIVE agent	10 hw
BN04	-daq/-dak	SIMILARITY / belonging	-
BN05	-siz/-suz	WITHOUT (privative)	-
BN06	-la/-le	VERBALIZER (noun→verb)	-

▼

Layer 3 — AA: Vowel Pattern (17 codes) | BI: Case System (6 codes)

AA VOWEL PATTERNS — 17 codes (25,304 tokens)

Vowels sit ON the consonant skeleton — same root, different vowels = different meaning:

Code	R1	R2	Function	Tokens
FA3L	a	0	Minimal noun	5,243
FA3AL	a	a	Open/action noun	7,565
FA3IL	a	i	Engaged noun	3,925
FU3UL	u	u	Deep/internal	1,283
FI3AL	i	a	Broken plural	733
+ 12 more vowel codes including 5 broken plural patterns

كِتاب (kitab) = FA3AL pattern on ك-ت-ب root
كاتِب (katib) = FA3IL on same root = different meaning

BI CASE SYSTEM — 6 codes (BC01-BC06)

BI has NO internal vowel pattern — uses SUFFIX stacking:

Code	Suffix	Case	Function
BC01	(bare stem)	NOMINATIVE	Subject (default)
BC02	-ni/-nı	ACCUSATIVE	Direct object
BC03	-ga/-ge/-qa	DATIVE	Direction / recipient
BC04	-da/-de/-ta	LOCATIVE	Location (at/in)
BC05	-dan/-den/-tan	ABLATIVE	Source (from)
BC06	-ning/-nıng	GENITIVE	Possession (of)

at (horse) → at-nı (horse-ACC) → at-qa (to-horse) → at-da (at-horse)

▼

Layer 4 — Grammar (external markers)

AA GRAMMAR — 11 verb defs + 13 noun defs

Dimension	Values	Marker
TENSE	MADI (past) \| MUDARI (present) \| AMR (command)	prefix/suffix
PERSON	1st \| 2nd \| 3rd	ي/ت/ن/أ prefixes
NUMBER	SINGULAR \| DUAL \| PLURAL	suffix
GENDER	MASC \| FEM	ة suffix
CASE	MARFU (nom) \| MANSUB (acc) \| MAJRUR (gen)	ending vowel
DEFINITENESS	DEFINITE (ال) \| INDEFINITE (tanwin) \| CONSTRUCT	prefix/suffix

Verb grammar: 15,333 tokens coded | Noun grammar: 34,058 tokens coded

BI GRAMMAR — 10 codes (BG01-BG10)

Code	Category	Suffix	Function
BG01-03	POSSESSIVE	-im/-ing/-i	My / Your / His
BG04-06	POSSESSIVE PL	-imiz/-ingiz/-lari	Our / Your / Their
BG07	VOICE	-il/-in/-ul	PASSIVE
BG08	NEGATION	-ma/-me	NOT
BG09	QUESTION	-mu/-mi	? (interrogative)
BG10	VOWEL HARMONY	(structural law)	Back↔Front

▼

Layer 5 — BI Compound Formation (8 rules) | AA: N/A (uses idhafa)

BI COMPOUNDS — 8 codes (COMP01-COMP08)

Code	Pattern	Function	Example
COMP01	NOUN + NOUN	Descriptive (modifier first)	qara tas = black stone = pencil
COMP02	VERB + -GAN	Participial compound	qorugan = protected place = fortress
COMP03	NOUN + VERB	Incorporating compound	basmaq = to stamp/print
COMP04	NUM + NOUN	Numerical compound	uch oq = three arrows (tribal name)
COMP05-08	Reduplicated, Echo, Copulative, Denominative

BI word order: MODIFIER + HEAD (opposite of AA idhafa where HEAD + MODIFIER)

Cross-Reference — The Diwan Bridge

How AA and BI connect through Kashgari's Diwan

Kashgari provides the AA gloss for each BI headword. AMR AI then:

1. Takes the BI headword → decomposes via BV/BN/BC/BG codes
2. Takes Kashgari's AA gloss → traces each Arabic word to its ROOT
3. Each root → letter-level semantic analysis via amr_alphabet
4. Each root → tasrif code (L1 consonant + L2 vowel + L3 grammar)
5. Each root → Quranic token count (proof of existence, not from weights)

Result: Every BI word connects through its AA gloss to Quranic roots, with full letter/root/tasrif analysis on both sides. No flat strings. No weight contamination. Every claim traceable to 28 letters and 77,877 tokens.

Example — Enriched Entry vs Flat String

DW891 — بكو (buko) = الملك والحاكم والعاقل

BEFORE:

aa_gloss = "al-malik wa-l-hakim wa-l-aqil"

AFTER:

BI SIDE:
  Headword: بكو (buko)
  Root: ب-ك [22] = CLOSURE(2) + TAP(20)
  Structure: BC01 (bare stem, nominative)

AA SIDE — 3 roots from Kashgari's gloss:

  الملك — م-ل-ك [90] = ENCLOSURE + CONNECTION + TAP
    L1: BASE | L2: FA3IL | L3: SINGULAR MASC DEFINITE
    Quranic: 185 tokens | Q1:4 مَٰلِكِ يَوْمِ الدِّينِ

  الحاكم — ح-ك-م [68] = CONSTRICTION + TAP + ENCLOSURE
    L1: A_BETWEEN_R1R2 | L2: FA3IL | L3: SINGULAR MASC DEFINITE
    Quranic: 195 tokens | Q2:32 ٱلْحَكِيمُ

  العاقل — ع-ق-ل [200] = DEPTH + STRIKE + CONNECTION
    L1: BASE | L2: FA3IL | L3: SINGULAR MASC DEFINITE
    Quranic: 49 tokens | Q2:44 تَعْقِلُونَ

Data Flow Summary

BI HEADWORD → bitig_tasrif decompose → BI root + suffix codes
↓
KASHGARI'S AA GLOSS (from Arabic MS, not Dankoff)
↓
Each AA word → amr_tasrif → L1 (consonant) + L2 (vowel) + L3 (grammar)
↓
Each AA root → amr_alphabet → letter semantics + abjad + positional behaviour
↓
Each AA root → quran_word_roots → token count (existence proof from 77,877 tokens)
↓
ENRICHED ENTRY — every claim traces to letters and Quranic tokens. Zero weights.

AMR AI — Letter / Root / Tasrif Architecture

How AA (ORIG1) and BI (ORIG2) words are structured, analyzed, and cross-referenced