ABCDE / アベケデ [a.be.ke.de] is a small language for connection and mutual aid:
eight semantic atoms, five core letters, no inflection. Numbers and external
labels are auxiliary. This is a review draft, not a released v1.
- Core: the grammar in this document and the atom/marker inventory in language.json. The inventory tables below are generated.
- Conventions:
conventions,legacy, andtime_conventionin that same JSON, rendered as dictionary.md. They specialize meaning, never overwrite a parse. - Examples: examples.json, rendered as examples.md. Natural translations state their assumptions.
- Grounding: actual people, objects, places and shared context. The interpreter does not supply these.
The README is introductory; adaptations.md is non-normative.
The Python code implements grammar rather than defining a second lexicon.
python3 tools/build_docs.py --check detects stale generated text; conformance
fixtures test the prose's grammar decisions. Change rules and tests together.
Passing tests does not establish real-world communicative adequacy.
| Atom | IPA | Core field | 意味の核 |
|---|---|---|---|
A |
[a] | motion/change | 動き・変化 |
E |
[e] | relation | 関係 |
BA |
[ba] | self/source | 自己・起点 |
BE |
[be] | other/target | 他者・対象 |
CA |
[ka] | entity/matter | 存在・物 |
CE |
[ke] | positive/desirable | 肯定的・望ましい |
DA |
[da] | action/power | 行為・力 |
DE |
[de] | negative/undesirable | 否定的・望ましくない |
Atoms name semantic fields, not English/Japanese dictionary words or truth values.
Bare A does not mark future; bare E does not mark past. CE/DE are evaluations
relative to an unspecified standard, not Boolean truth values. Within compounds,
BA/BE contribute self/source and other/target; independent participant readings
“I” and “you” belong to conventions.
Tokenization is unique: A, E, or B/C/D followed by A/E. Join one or more atoms
without separators. The last atom is the head, and all earlier atoms form an
ordered, flat list of modifiers of that head. There is no recursive dictionary
substitution or hidden grouping. The structural gloss is head〈modifier + …〉.
ADACA parses as [A, DA] → CA, an entity related to motion and action. “Water”
is a survival convention, not a logical consequence. CECADA parses as
[CE, CA] → DA; “eat” is a whole-word convention, not dictionary substitution
of CECA into a compound tree. Unknown compounds retain their atoms and head.
Repeating an atom reinforces its field: CECE is intensified positive evaluation;
DEDE is intensified negative evaluation. It introduces no new size scale.
Mixed positive/negative modifiers are retained without Boolean simplification.
Space-separated repeated words are separate terms, not a plural operator.
Use a count when known (3BACA); otherwise plurality is unspecified.
Outside brackets use uppercase A B C D E, ASCII digits and spaces, and - ! ? [ ]
only in their defined forms. One nonempty line is one sentence or fragment.
LF/CRLF are accepted, blank lines ignored, repeated spaces equivalent to one.
Tabs, lowercase and other punctuation outside brackets are invalid. A line is
never split by guessed semantics.
A term is an optional attached DE-, optional attached count, then a word or
external label. A count is 0 or an unsigned decimal integer without leading
zeros. Bare numerals are invalid. Counts mean counts of the referent, not
inferred dimensions: 5[egg] does not mean the volume of five eggs.
External labels are nonempty text in one matching bracket pair: [Tokyo], [山田],
[New York], 5[egg]. Preserve spaces and spelling inside. Empty/whitespace-only
labels, control characters, nesting and multiline labels are invalid. They are
opaque entities, not automatically people, verbs or translated words. They
cannot be predicates or concatenate with atoms. DE-[Tokyo] negates the whole label.
Day (ACA) and step (BADA) are natural-unit conventions, not calibrated measures.
The interpreter computes no physical lengths or calendar dates. Numerals may use
a mutually understood local digit reading. V0's digit pronunciations remain an
optional adaptation; eight semantic atoms does not mean eight total spoken forms.
line := [mood] [tense] (clause [E reference] | fragment)
clause := subject predicate object
| subject predicate
| predicate object
| predicate
fragment := term that cannot be a predicate
Brackets in this grammar mean optional. Every position is one term, separated
by spaces. A predicate is an unquantified non-external word headed by DA,
CE or DE; it may have DE-. These fields express action/power or evaluation.
Any term may fill another role, subject to the reserved bare E rule below.
This is deliberately restricted grammar, not a parser for all v0 prose.
- Three body terms mean subject → predicate → object/target (S-P-O). Roles
never change based on plausibility or lexical lookup.
BA CEDA BEandBE CEDA BAdescribe different helpers and beneficiaries. - With two terms, retain every eligible S-P and P-O parse.
CE DEpermits both.CEDA BApermits only P-O becauseBAis not a predicate. - One predicate has unspecified subject and object. One non-predicate is a fragment; no proposition is invented.
- Other shapes are invalid. Missing predicates are never repaired.
- An omitted role is
null, never automaticallyBAorBE. An unspecified object may be unnecessary for an intransitive reading or may need context; the interpreter does not invent a valency distinction.
Word order matters. No fronting, multiword noun phrases, implicit possession, coordination, or extra place/time slots are provided. Compound modifiers must be joined inside a word. An external label is also one term. Do not silently combine adjacent terms to repair a sentence.
| Marker | Meaning | Scope |
|---|---|---|
E? |
question | whole sentence / 文全体 |
A! |
emphasis | whole sentence / 文全体 |
A- |
future | whole sentence / 文全体 |
E- |
past | whole sentence / 文全体 |
DE- |
negation / 否定 | one attached term / 直結する1語 |
At most one mood, then at most one tense, precedes the body; order is mandatory. Duplicate, conflicting and misplaced markers are invalid. No tense marker means unmarked time (a present/general reading is possible, not asserted). Past or future does not encode completion, certainty or ability.
DE- negates one entire term, including count: DE-3BACA means “not three persons”
under the person convention, not “three non-persons”. On a predicate it denies
that predication for the participants. It is not a clause-negation particle.
DE-CE differs from lexical DECE. Repeated DE- is invalid; negation alone
never means “do not”, “cannot”, or danger.
A! adds emphasis, not imperative mood. A survival phrase may read an emphatic
action as an appeal (“Help me”, “Do not eat”), but must state that convention
and preserve the emphatic-assertion reading in its structural output. Bare
CE/DE do not encode context-free yes/no answers. Only whole-line CE! and
DE! expand to A! CE and A! DE. Bare E? has no body and is invalid here.
Standalone E is reserved for one clause suffix E reference. The reference
is exactly one term after the body, not another object. The relation attaches
to the whole clause; its type is unspecified, not automatically after, before,
with, possession, or causation. E alone may be inspected as an atom fragment;
otherwise it is not a bare participant or reference. Compounds containing E
remain ordinary words.
In the survival profile, future/past + E + counted ACA may mean that many days after/before the utterance. Bare E itself means only relation. Other readings remain possible.
生存会話の慣用では未来・過去 + E + 数付き ACA を発話時点からの日数差として読める。裸の E 自体は関係のみ。他の読みも残る。
A- EADA E 3ACA has an omitted subject, a return convention, and a possible
“three days after the utterance” reading. Who returns and the return destination
remain unspecified. BA EADA E 3ACA supplies no time direction. E- BA EADA E 3ACA
offers a past offset. Negated references never activate the time convention.
The broader ACA readings remain visible.
python3 -m abcde and abcde.analyze(text) provide the same analysis. JSON has
a language version, aggregate status and one result per nonempty line. A bad
line is invalid; valid lines in the same input remain available. Output includes:
terms (atoms, ordered modifiers, head, count, negation, bilingual literal),
markers, all possible role maps, relation, conventional readings and diagnostics.
Role/reference indices address the terms array. Fragments have no role map.
ambiguous specifically means multiple parses or registered lexical readings.
valid means structurally accepted, not semantically complete. Omissions,
external labels, unknown compounds and pragmatic uncertainty are separate
notices. Legacy warnings identify retired readings, never activate an old
parser or automatically migrate text.
Base-word candidates are explicitly before count and negation. The literal,
negated and count fields carry these operators; never show a positive base
reading as the translation of a negated word. Natural whole-sentence candidates
are only available for registered examples, with assumptions. Generic input
gets structured bilingual glosses and word candidates, not invented fluent prose.
--core-only suppresses word, phrase and time conventions while preserving
parses, omissions and legacy notices. Reverse --from ja / --from en uses
exact phrasebook matching, never free-text translation. Outer/repeated whitespace
is normalized; English matching is case-insensitive. Punctuation, negation,
names and extra words are not discarded. Unsupported input gives unsupported
and no matches. Supported matches include a fresh forward analysis.
In the example data, candidate: false marks analysis-only illustrations;
they are documented but never offered as natural sentence translations.
Python 3.9+ and its standard library suffice; there are no network calls or runtime dependencies. The implementation limits input to 8192 characters per analysis (a resource bound, not a linguistic rule). CLI exit status: 0 for accepted/ambiguous input, 2 for invalid/unsupported input or CLI misuse.
C is [k]. Stress the final atom, with audible word boundaries and a clear
pause between sentences. Ordinary atoms have short vowels; A-, E-, DE-
have extended vowels. A! and E? are short marked utterances. These guidelines
are not a validated speech-recognition protocol.
This draft deliberately leaves referents, the kind of help, and many action details to context. It does not express arbitrary Japanese/English or every linguistic distinction. A larger grammar and an ABCDE-native learner are outside this pass. See migration.md for compatibility and review decisions.