Unicode & IPA¶

ELANG identifies sounds with IPA strings, and it is strict about exactly which characters are allowed and how they combine. Behavior is pinned so it does not drift with the host environment.

Pinned sources¶

• Unicode: pinned to 18.0.0 for all NFC and General-Category behavior. The reference implementation generates its tables from the official UCD 18.0.0 and does not use host-OS Unicode behavior.
• IPA chart: pinned to the IPA 2026 chart issue.
• IPA Repertoire v1: a narrow, explicit allowlist of approved base symbols, combining diacritics, and spacing modifiers. Anything outside the allowlist is rejected.

The repertoire is distributed as strict UTF-8 JSON (no BOM), canonicalized with RFC 8785 JCS, and verified against a pinned SHA-256 digest shipped with the validator.

Rejected characters¶

Every ELD string must be NFC; non-NFC input is rejected rather than normalized. In addition, the following are rejected in every ELD string:

• Unicode control characters (Cc) and format characters (Cf),
• private-use characters (Co),
• unassigned code points (Cn),
• all Unicode noncharacters (including U+FDD0–U+FDEF and the plane-end FFFE/FFFF code points),
• surrogates and any invalid UTF-8.

Standalone stress marks, syllable-boundary markers, and tone notation are excluded from phoneme and allophone identifiers; they are reserved for future prosody features.

Logical units¶

An IPA identifier is measured two ways, both enforced independently:

• 1–8 Unicode scalar values, and
• at most 4 logical IPA base units.

A combining diacritic or a spacing modifier must attach to a valid base symbol. A standalone modifier, or a combining mark with no preceding base, is invalid.

Each logical base unit allows at most one modification — either one combining diacritic or one spacing modifier, never both. NFC-precomposed characters are interpreted by their canonical decomposition when counting base-plus-modifier units.

Tie bars¶

The tie bars U+0361 (upper) and U+035C (lower) join exactly two IPA base symbols into one logical base unit. Tie bars cannot overlap or chain. A modifier for a tied unit may appear only after the second base. Multiple independent tied units, and a mix of tied and untied units, are allowed.

Identity and uniqueness¶

• Untied tʃ and tied t͡ʃ are distinct IPA identities.
• t͡ʃ (upper tie bar) and t͜ʃ (lower tie bar) are equivalent for uniqueness — they count as the same identifier.

The authored tie-bar code point is preserved in the source and in canonical output; the upper/lower distinction is folded only when computing comparison keys. There is no other phonetic equivalence beyond Unicode canonical equivalence under strict NFC and this explicit tie-bar equivalence. IPA uniqueness is global across all root and allophone declarations, and there is no case folding for IPA.

Examples¶

These outcomes assume the final approved allowlist; they illustrate the structural rules.