ELD — phoneme inventory¶

.eld is the phoneme inventory, written in KDL 2.0. It is the one linguistic module ELANG v0.1.0 defines. An ELD document describes a language's sounds as groups of root phonemes, each with a romanization, optional allophones, and optional conditional realization rules.

Example¶

phonemes {
    consonants {
        phoneme "t" {
            roman "t"
            allophone "ɾ"
            allophone "tʰ" {
                roman "th"
                when {
                    word-position "initial"
                }
            }
        }
    }
    vowels {
        phoneme "a" {
            roman "a"
        }
    }
    diphthongs {
        phoneme "aɪ" {
            roman "ai"
        }
    }
}

Top level¶

An ELD document contains exactly one top-level phonemes node with an explicit, nonempty block and zero arguments or properties. Empty, whitespace-only, or comment-only files are rejected, as are any additional top-level nodes.

Groups¶

phonemes {
    consonants { /* ... */ }
    vowels     { /* ... */ }
    // optional custom groups:
    diphthongs { /* ... */ }
}

• Exactly one consonants group and one vowels group, each with ≥ 1 root phoneme.
• Up to 16 groups total (so up to 14 custom groups); every custom group must be nonempty.
• Group names are unique. Custom group names match the regex [a-z]+(?:-[a-z]+)* and are 1–32 ASCII characters. consonants and vowels are reserved.
• Group nodes take zero arguments/properties and require a child block whose only direct children are phoneme nodes.
• Group source order carries no meaning. The canonical order is consonants, vowels, then custom groups in ASCII bytewise order.

Root phonemes¶

phoneme "<IPA>" {
    roman "<spelling>"
    // optional:
    when { /* ... */ }
    // zero or more:
    allophone "<IPA>"
}

• phoneme takes exactly one IPA string argument, zero properties, and a mandatory nonempty block.
• Exactly one roman child; optional zero-or-one when; zero to 16 direct allophone children.
• At most 256 root phonemes across all groups.
• Root IPA identifiers are globally unique across every group and allophone. You cannot "reuse" a sound by placing it in another group.
• Only roman, when, and allophone children are allowed. frequency, distribution, description, and any other field are rejected.
• Canonical child order: roman, then optional when, then allophones sorted by IPA. Source declaration order carries no priority.

Allophones¶

An allophone is a conditional surface pronunciation of its root.

// Fieldless: a bare leaf (no braces).
allophone "ɾ"

// With fields: a nonempty block.
allophone "tʰ" {
    roman "th"
    when { word-position "initial" }
}

• allophone takes exactly one IPA string argument and zero properties, and must be a direct child of a root phoneme (never nested under another allophone).
• Optional zero-or-one roman; optional zero-or-one when.
• A fieldless allophone is a bare leaf — allophone "ɾ". Writing allophone "ɾ" {} with an empty block is invalid. When fields are present, the block must be nonempty.
• An omitted roman inherits the parent root's exact-case romanization. Inherited romanizations do not count as explicit duplicates.
• An omitted when means unconditional eligibility. An explicit empty when {} is invalid.

Independence of conditions¶

A root's when and an allophone's when are evaluated independently — the child does not inherit the parent's condition. Multiple pronunciations can be eligible at once; the generator selects among them with no weighting or declaration priority. If none is eligible in a given context, the inventory is still valid — the generator simply avoids that placement or chooses a different root. ELANG defines no fallback pronunciation.

Romanization¶

• roman is a scalar leaf: exactly one string argument, no properties, no block.
• An explicit value matches [A-Za-z]{1,8} — 1 to 8 ASCII letters, preserving case exactly in canonical output.
• Every explicit roman across all roots and allophones is globally unique using ASCII case-insensitive comparison. For example, th and TH collide. Inherited romanizations are exempt.

IPA identifiers¶

Every phoneme and allophone IPA string is validated against the pinned IPA Repertoire v1 and a strict set of Unicode and structural rules (logical units, modifiers, tie bars, length). Those rules are detailed in Unicode & IPA.

Canonical form¶

The canonical ELD serializer fixes group/root/allophone ordering, child ordering, condition ordering, indentation, line endings, and string quoting, so that semantically equal inventories always produce identical bytes.