greek_romanizer 0.2.1

Configurable Greek-to-Latin romanization: ASCII ELOT 743/ISO 843 style, plus English and German phonetic respellings, from one overridable rule set.

greek_romanizer

Configurable Greek → Latin romanization. One context-aware engine, three presets, every letter choice overridable.

import 'package:greek_romanizer/greek_romanizer.dart';

void main() {
  final official = GreekRomanizer.official();
  print(official.romanize('ευχαριστώ').text); // efcharisto

  final english = GreekRomanizer.englishReader();
  print(english.romanize('ευχαριστώ').text); // efcharistó

  // "efharisto" reads easier for an English speaker than "efcharisto":
  final easier = GreekRomanizer.englishReader(chiLetter: 'h');
  print(easier.romanize('ευχαριστώ').text); // efharistó
}

Presets

Preset	For	Notes
GreekRomanizer.official()	ASCII, official-style transliteration	close to ELOT 743 / ISO 843 (Greek passports, road signs), no accent marks
GreekRomanizer.englishReader()	an English speaker	phonetic respelling — collapses vowel digraphs, voices μπ/ντ/γκ at word start
GreekRomanizer.germanReader()	a German speaker	not a copy of englishReader — β, ζ, ξ, σ and γ's front/back split all diverge, because German reading habits differ from English ones

Why official carries no accents by default. Real ELOT 743 output on a passport or road sign doesn't show stress. Every romanizer still returns RomanizedText.stressed — a list of StressSpan offsets into the plain text — so a caller who wants to highlight the stressed syllable can, even without a rendered accent mark. Pass showStress: true on a preset (via copyWith, see below) to render the accent instead.

Why official is orthographic and the readers are phonetic. ELOT is letter-for-letter: Πειραιάς transliterates to Peiraias, which an unprimed reader says wrong — the word is pronounced pireas. The reader presets respell for pronunciation instead (Pireás), at the cost of no longer matching the passport spelling.

Overriding the rules

Every letter choice is a field on RomanizationSpec. Each preset factory exposes the fields that are safe to tweak without breaking the scheme's own identity:

final GreekRomanizer easier = GreekRomanizer.englishReader(chiLetter: 'h');

For full control, build a RomanizationSpec from scratch or copyWith a preset's spec and hand it to GreekRomanizer.custom:

final spec = RomanizationSpec.englishReader.copyWith(
  xLetter: 'x',       // read ξ as x instead of ks
  showStress: false,  // drop accent marks, keep the phonetic letter choices
);
final GreekRomanizer custom = GreekRomanizer.custom(spec);

Whole-word overrides

For loanwords or names the rule table gets wrong, skip the engine for that word entirely:

final romanizer = GreekRomanizer.official(
  wordOverrides: {'γιαούρτι': 'yaourti'},
);
romanizer.romanize('γιαούρτι').text; // yaourti

Matching is case-insensitive on the Greek key; the override's leading letter is re-cased to match the input (Γιαούρτι → Yaourti).

Rule reference

Vowels

Greek	official (literal)	reader (phonetic)
α ε η ι ο ω	a e i i o o	a e i i o o
υ (alone)	y	i
αι	ai	e
ει, οι, υι	ei, oi, yi	i
ου	ou	u

αυ / ευ / ηυ are context-sensitive in every preset: voiced (av / ev / iv, using vLetter) before a vowel or a voiced consonant (β γ δ ζ λ μ ν ρ), voiceless (af / ef / if) before a voiceless consonant (θ κ ξ π σ τ φ χ ψ) or at word end.

αύριο → ávrio      αυτό → aftó
Ευρώπη → Evrópi    ευχαριστώ → efcharistó

Consonants

Greek	Field	official	englishReader	germanReader
β	vLetter	v	v	w
ζ	zLetter	z	z	s
ξ	xLetter	x	ks	x
χ	chiLetter	ch	ch	ch
θ	thetaLetter	th	th	th
ψ	psiLetter	ps	ps	ps
δ	deltaLetter	d	dh	dh
γ (before ε ι η υ αι)	gammaFrontLetter	g (never splits)	y	j
γ (elsewhere)	—	g	gh	gh
σ / ς	doubleIntervocalicSigma / collapseDoubleSigma	s	s (σσ → s)	s / ss between vowels

German doubles a single σ between two vowels (its own s would otherwise read as /z/ there); English instead collapses an existing σσ pair down to one s. The two flags are mutually exclusive.

Nasal + stop clusters — verified against the official ELOT 743 / BGN-PCGN reference table (see Sources): μπ is the only one of the three that collapses to a bare voiced stop at word start and word end, in every preset (μπύρα → byra; a word-final μπ essentially never occurs in native Modern Greek vocabulary, but the official table specs it as one row alongside word-initial, so the engine honors it too). ντ and γκ stay literal in official everywhere, initial or medial (ντομάτα → ntomata, γκολ → gkol) — not symmetric with μπ, despite all three being the same phonological pattern. The reader presets collapse all three the same way at word start (not word end — that generalization isn't sourced, only the μπ row is), reasoned from Modern Greek pronunciation rather than pulled from the table:

Greek	word-initial, official	word-initial, readers	medial, official	medial, readers
μπ	b	b	mp	mb
ντ	nt	d	nt	nd
γκ	gk	g	gk	ng

No word-initial/medial split, same in every preset:

Greek	official	readers
τζ	t + zLetter	d + zLetter
τσ	ts	ts
γγ	ng	ng
γχ	n + chiLetter	n + chiLetter
γξ	n + xLetter	n + xLetter

τζ, γχ and γξ route through the same zLetter / chiLetter / xLetter fields as standalone ζ / χ / ξ, rather than hardcoding a letter — this matters in practice: a German phrasebook spells τζ as ds, not dz (German z reads as /ts/, so only a letter that reads as /z/ before a vowel — its own respelling of ζ — produces the right sound), which is exactly what germanReader's zLetter: 's' now produces automatically.

Stress

Modern Greek monotonic orthography carries the stress — the engine reads it, never guesses it. A tonos vowel (ά έ ή ί ό ύ ώ) always produces a StressSpan at that offset, whether or not the scheme renders the accent mark itself. In a vowel digraph the tonos sits on the second character, and the accent lands on the single resulting Latin vowel in a phonetic preset (καλοί → kalí). A diaeresis breaks a digraph (παϊδάκια, not *pedákia). πότε (when) and ποτέ (never) differ only by stress — dropping it teaches the wrong word.

A diaeresis and a tonos can land on the same vowel (ΐ, ΰ) when a stressed syllable is also the one broken out of a digraph, e.g. ταΐζω (to feed) or Ταΰγετος (Taygetus). These still carry a StressSpan — official ELOT 743 renders them as the base letter plus a combining acute (ḯ = U+00EF + U+0301), which showStress: false strips down to a plain vowel like every other accent.

Design notes

• Pure Dart, zero dependencies, no I/O — safe to call from an isolate.
• The engine returns stress offsets (RomanizedText.stressed), never a marked-up string. Nothing has to parse the output looking for accents.
• One RomanizationEngine, parameterized by RomanizationSpec, drives all three presets — no scheme derives its output from another scheme's string, so none of them can lose information the others have.

Sources

• ELOT 743 / ISO 843, cross-checked letter-by-letter against the official UK Government BGN/PCGN reference table for Greek (ROMANIZATION_OF_GREEK.pdf, checked for validity and accuracy 2017), the Greek government's own converter at passport.gov.gr, and the UN romanization working group's Greek report. That cross-check is what caught the μπ word-final case and the ΐ/ΰ stress-and-diaeresis case.
• The reader presets are standard Modern Greek phonology, reasoned directly rather than pulled from a converter (no official standard covers "readable for an English/German speaker"). German-specific choices (β → w, ζ/τζ → s/ds) are additionally corroborated by a native German-language phrasebook, Wikivoyage's Sprachführer Griechisch. DIN 31634, Germany's own Greek transliteration standard, was checked and ruled out as a source for this — it's a scholarly, accent-stripping system built for library cataloging of Ancient/Byzantine Greek (β → b, not v or w), not a Modern Greek pronunciation guide.